Split (1)

train · 5.12k rows

playlist stringclasses 160 values	file_name stringlengths 9 102	content stringlengths 29 329k
MIT_RES9003_Brains_Minds_and_Machines_Summer_Course_Summer_2015	Lecture_41_Shimon_Ullman_Development_of_Visual_Concepts.txt	The following content is provided under a Creative Commons license. Your support will help MIT OpenCourseWare continue to offer high-quality educational resources for free. To make a donation or view additional materials from hundreds of MIT courses, visit MIT OpenCourseWare at ocw.mit.edu. SHIMON ULLMAN: What I'm going to talk about today is very much related to this general issue of using vision, but with sort of the final goal of seeing and understanding, in a very deep and complete manner, what's going on around you. And let me start with just one image we've all-- we see images all the time, and we know that we can get a lot of meaning out of them. But just, you look at one image, and you get a lot of information out of it. You understand what happened before, that was some kind of a flood, and why these people are hanging out up there on the wires, and so on. So all of this, we get very quickly. And what we do, usually, in computational vision in much of my own work is to try to understand computational schemes that I can take as an input, an image like this, and get all this information out. But what I want to talk about in the first part-- I'm going to break the afternoon into two different talks on two different topics, but they are all closely related to this ultimate goal which drives me, which is, as you will see-- I think it will become apparent as we go along-- this sort of complete and full understanding, and complicated concepts, and complicated notions that we can derive from an image. But what I'm going to discuss in the first part is sort of how it all starts. And this combines vision with another topic which is very important in cognition-- part of CBMM as well. But it's not just this particular center, it's part of understanding cognition-- is infant learning and how it all starts. And certainly for vision, and for learning, this is a very interesting and fascinating problem. Because here, you think about babies, they open up their eyes. And they see-- they don't-- they cannot understand the images that they see. You can think that they see pixels. And they watch the pixels. And the pixels are transforming. And they look at the world as things change around them and so on. And what this short clip is trying to make explicit is, somehow, these pixels, over time, acquire meaning. And they understand the world, what they see. The input changes from changing pixels and light intensities into other infants, and rooms, and whatever they see around them. So we would like to be able to understand this and to do something similar. So to have a system, imagine that we have some kind of a system that starts without any specific world knowledge wired into it. It watches movies for, say, six months. And at the end of the six months, this system knows about the world, about people, about agents, about animals, about objects, about actions, social interactions between agents the way that infants do during the first year of life. So the goal isn't-- for me at least, it's not really-- the interest is not necessarily to engineer, the engineering part of it to really build such a system, but to think about and develop schemes that will be able to deal with it and do something similar. I think it's also-- maybe I'll mention it at the end-- it's also a very interesting direction for artificial intelligence to think about not generating final complete systems, but generating baby systems, if you want, that have some interesting and useful initial capacities. But the rest is just, they watch the world, and they get intelligent as time goes by. I'm going to talk initially about two particular things that we've been working on and we selected because we thought it particularly interesting. And one has to do with hands, and the other one has to do with gaze. And the reason that we selected them is, I'll show in a minute, visually, for computer vision, being able to deal with hands of people's in images and dealing with direction of gaze are a very difficult problem. And a lot of work has been done in computer vision dealing with issues related to hands and to gaze. They're also very important for cognition in general. Again, I will discuss it a bit more later. But understanding hands and what they are doing, interacting with objects, manipulating objects, action recognition-- so hands are a part of understanding the full scheme, the whole domain of actions. And social interactions between agents is a part of it. Gaze is also very important for understanding intentions of people, understanding interactions between people. So these are two basic-- very basic type of objects or concepts that you want to acquire that are very difficult. And the final surprising thing is they come very early in infant vision, one of the first things to be learned. So you can see here, when I say hands are important, for example, for action recognition, I don't know if you can tell what this person is doing, for example. Any guess? Yeah, you say what? What is he doing? AUDIENCE: Talking on the phone. SHIMON ULLMAN: Talking on the phone. And we cannot really see the phone. But it's more where the hand is relative to the ear and so on. And you know, we can see the interactions between agents and so on. A lot depends on understanding the body posture, and in particular, the hands. So they are certainly very important for us. I mentioned that they are very difficult to be able to automatically be able to extract them and localize them in the image. And there are two reasons for this. One is that hands are very flexible, much more so than most rigid objects we encounter in the world. So a hand does not have a typical appearance. It has so many different appearances that it's difficult to handle all of them. And the other reason is that hands in images, although they are important, very often, there is very little information in the image showing the hand. Just because of resolution size, partial occlusion, we know where the hands are. But we can see very little of the hands. We have the impression here, when we look at this, we know what this person is doing, right? He's holding a camera and taking a picture. But if you take the image-- you know, where the hand and the camera are-- you know, this is the camera, this is the hand, and so on-- it's really not much information. But we can use it very effectively, and similarly in the other images that you see here. Children, or infants, do this ability to deal with hands-- and later on, we'll see, with gaze-- in a completely unsupervised manner. Nobody teaches them, look, this is a hand. Even it cannot be even theoretically possible, because this capacity to deal with, say, hands and gaze comes at the age of three months, way before language starts to develop. So all of this is entirely unsupervised, just watching things happening in an unstructured way and mastering these concepts. And when you try to imitate this in computer vision systems, there are not too many computer vision system, learning system that can deal well with unsupervised data. And I can tell you without going-- I don't want to elaborate on the different schemes that exist and so on. But the kind of thing that exists cannot-- nobody can help, nobody can learn hands in an unsupervised way. It may be interesting to know, just anecdotally, when we actually started to work with this, deep networks were not exactly what they are today. If you go back in the literature, and we see when the term "deep networks" and the initial work on deep networks started by-- at least in the group of Geoff Hinton. Yann LeCun was doing independent things separately. But the goal was to learn everything in an unsupervised way. They were labeled as, that's the goal of the project, to be able to build a machine. And the machine, you will not need supervision. You just aim it at the world, and it will start to absorb information from the world and build a internal representation in a completely unsupervised way. And they demonstrated it on simple examples-- for example, on MNIST on digits, that you don't tell the system that there are 10 digits and so on. You just show it data, and it builds a deep network that automatically divides the input into 10 different classes. And in interesting ways, it also divides subclasses. There is a closed 4, and an open 4, and so on with something very natural. And it was an example of dealing with multiple classes in an unsupervised manner. But when you try to do something like hands, which we tried, I mean, there is just-- it basically failed as an unsupervised method. And the problem remained difficult. Here is a quote for Jitendra Malik, who is-- those of you who deal with computer vision would know the name. He's sort of a leading person in computer vision. "They say that dealing with body configuration in general, and hands in particular, is maybe the most difficult recognition problem in computer vision." I think that's probably going too far. But still, you can realize that people took it as a very difficult problem. On the unsupervised way, which is still a big open problem, the biggest effort so far has been a paper-- it's already a few years ago-- a collaboration between Google and Stanford by Andrew Ng others in which they took images from 10 million YouTube movies. And they tried to learn whatever they could in an unsupervised way. And they designed a system that was designed to get information out in an unsupervised manner. And basically, they managed, from all this information, to get out three concepts what happened in the machine that you-- it developed units that were sensitive, specifically, to three different categories. One was faces, the other one was cats. That's from their paper. It's not easy to see the cat here, but there is a cat. They found the cat. And there is a sort of a torso, upper body from-- upper body. Three concepts-- after all of this training, three concepts sort of emerged. And in fact, only one of them, faces, was really there, that there were units that were very sensitive to faces. For the other cases, like cats and upper bodies, it was not all that selective. And by the way, cats is not very surprising. You know why cats came out in these movies? If you watched YouTube you would know, literally, it's also the most salient thing. After faces, it's literally the case that if you take random movies, or millions of movies from YouTube, many, many of them will contain cats. So in the database, it was the most-- after faces and bodies, it was the third most frequent category. So it wouldn't do, hands or gaze and so on. It's really picking up only things which are very, very salient and very, very frequent in the input. And as I said, babies do it. And now people started to work to look more closely at it. One technique was to put a sort of webcam on infants. This is not an infant. This is a slightly older person, a toddler. But they do it with infants, and they look at what the babies are looking. And what babies are looking at the very initial stages are faces and hands. They really like faces. And they really like hands. And they recognize hands. And they know sort of what-- they have, already, information and expectation about hands in a very early age. So it's not just even the visual recognition that they group together images of hands, but they know that hands, for example, are the causal agents that are moving objects in the world. And this is for an experiment by Rebecca Saxe. There is a person here working with Rebecca Saxe, right? Did she talk already in the-- AUDIENCE: She will. SHIMON ULLMAN: She will. And she's worth listening to. So this is from one of her studies in which they showed infant-- these are slightly older infants. But they showed infants, on the computer screen, a hand moving an object. This is not taken from the paper directly. This, I just drew-- but a hand moving an object-- in this case, a cup or a glass. And the infant watches it for a while and sort of gets bored. And after they do it, they show the infant either the hand moving alone on the screen or the glass moving alone on the screen. And the hand moving alone, on itself, on the screen, they are not-- still bored. They don't look at it much. When the cup is moving on the screen, they are very surprised and interested. So they know it's the hand moving the cup. It's not the cup moving the hand or they have equal status, it's the motion of this configuration. The originator, or the actor, or the mover is the hand. So this is at seven months. But it has been known that this kind of motion that this one-- that an object can cause another object to move, this is something that babies are sensitive to, not only at the age of seven months, but it's something that appears in infants as early as you can imagine. And you can test. And this, for us-- I'll tell you what they are sensitive to. And for us, this was sort of the guideline, or sort of the door, the open door, how to-- what may be going on that lets the infant speak up specifically on hands and quickly develop a well-developed face detect-- hand detector. So infants are known to-- have been known to be sensitive-- they are sensitive to motion in general. They are really following moving objects. And they use motion a lot. But motion by itself is not very helpful if you want to recognize hands. It's true that hands are moving. But many other things are moving as well. So if you take random videos from children's perspective, they see doors opening and closing. They see people moving back and forth, coming by and disappearing. Hands are just one small category of moving things. But they're also sensitive, as I said, not just to motion but to this particular situation in which an object moves, comes in contact with another object, and causes it to move. And this is not even at the level of objects. At three months, they don't even have a well-set notion of-- they just begin to organize the world into separate objects. But you can think of just cloud of pixels if you want, moving around. They come in contact with stationary pixels and causing them to move. Whenever this happens, infants pay attention to this. They look at it preferentially. And it's known that they are sensitive to this type of motion, which called a mover event. And a mover event is this event that I just described that something is in motion, comes in contact with a stationary item, and causes it to move. So we started with developing a very simple algorithm that, all it does, it simply looks on video, on changing images, watching for, or waiting for, mover events to occur. And the way it's defined, it's very simple. In the algorithm, we divide the visual field into small regions. And we monitor each one of these cells in the grid for the occurrence of a mover, which means that there will be some optical flow, some motion coming into the cell, and then leaving the cell, and sort of carrying with it the content of the cell. So this does require some kind of optical flow in change detection. And all of these capacities are known that they're in place before three months of age. And now, look at an image of a person manipulating objects when all you do is simply monitor all locations in the image for the occurrence of this kind of a mover event. What you should see, or what you should pay attention to, is the fact that motion, by itself, is not particularly-- is not doing anything. What the algorithm is doing is, whenever it detects a mover event, it draws a square around the mover event and continues to follow it to track it for a little bit. So you can see, the minute the hand is moving something, it's detected as a hand moving things on their own, are not triggering the system here. The hand is moving. So this, by itself, is not the signal. But the interaction between the hand and an object is something that it's detected based on these very low-level cues. It doesn't know about hands. It doesn't know about objects. But you can create-- you can see, here, a false alarm, which was interesting. You can probably understand why, and so on. So here's an example of what if you just let it-- this scheme run on a large set of hours and hours of videos. Some of these videos do not contain anything related to people. Some of them, there are people, but just going back and forth, entering the room, leaving the room, but do not manipulate objects. Nothing specific happens in these videos. The system that is looking for these kinds of mover events finds very rare occasions in which anything interesting happens. But you can see the output that happened. This is just examples of output of just, from these many videos, the kind of images that it extracted by being tuned specifically to the occurrence of this specific event. You can see that you get a lot of hand. These hands are the continuation. The assumption here, which we, again, took and modeled after what-- when infancy, something that starts to move, they track it for about one or two seconds. So we also tracked it for about one or two seconds. And we show some images from this tracking. These are some false alarms. There are not very many of them. But these are examples of some false alarms, that something happened in the image that caused the scheme for the specific mover event to be triggered. And it collected these images. But on the whole, as it shows here, you get very good performance. It's actually surprising that, if you look at all the ima-- all the occurrences of hands touching objects, that 90% of them in all these collection of videos were captured by the image. And the accuracy was 65%. So there were some false alarms, but the majority were fine. And what you will end up with is a large collection which is mostly composed of hands. So now you have, without being supplied with external supervision-- here is a hand, here is a hand, here is a hand-- suddenly, you have a collection of 10,000 images. And most of them contain a hand. And now if you look at this, and you apply completely standard algorithms for object recognition, this is great for them-- this is sufficient for them-- in order to learn a new object. So if you do a deep network-- but you can do even simpler algorithms of various sorts-- what you will end up using this collection of images, which were identified as belonging to the same concept by all triggering the same special event of something causing something to move, you will get a hand example. So these are just-- I will not play them-- lots and lots of movies. Some of them can play with for half an hour without having a single event of this kind. Others are pretty dense with such events. And they are being detected. And what is shown here, it's not the greatest image. But all these squares here, the yellow squares, are-- having gone through this first round of learning these events, this is the output of a detector. You take the images that were labeled in this way. You give it to a standard computer vision classifier that takes these images and finds what's common to these images. And then on completely new images-- static images now-- this marks, in the image, following this learning, where the hands will be-- the hands are. Now, this is a very good start in terms of being able to learn hands, and not only learn hands. As I showed you before, very early on, the notion of learn is automatically a sort of-- in terms of the cognitive system, is closely associated with moving objects, causing objects to move, as we saw in the Rebecca Saxe experiment. This also happens in this system. But this continues to develop. Because eventually we can learn hands not only in this grasping configuration. Later on we want to be able to recognize any hand in any configuration. And so the system needs to learn more. And this is accomplished by something that, again, the details, or the specific application here, is less important than the principle. And the principle is sort of two subsystems in the cognitive system training each other. And together, by each one advising the other, they reach, together, much more than what any one of the systems would be able to reach on its own. And the two subsystems in this case, one is the ability to recognize hands by their appearance. I can show you just an image of a hand without the rest of it, the rest of the body or the rest of the image. And you know, by the local appearance, that this is a hand. You also know, here, you cannot even see the hands of this woman. But you know where the hands are. So you can also use the surrounding context in order to get where the hands are. So these are two different algorithms that are known in computer vision. They are also known in infants. People have demonstrated, sort of independently before we have done our work, that infants associate the body, and even the face-- when they see a hand, they immediately look up, and they look for a face. And they are surprised if there is no face there. So they know about the association between body parts surrounding the hand and the hand itself. And you can think about it as-- we saw this image before. Here the hand itself is not very clear. But you can get to the hand by-- if you know where the face is, for example, that you can go to the shoulder, and to the upper arm, and lower arm, and end up at the hand. So people have used this in computer vision as well, finding hands. They also use this idea of finding the hands on their own by their own appearance or using the surrounding body configuration. And the nice thing is, instead of just thinking of, here are two methods, two schemes that can both produce the same final goal, they can also, during learning, help each other and guide each other. If you think about it, the way it goes it is shown here, that, sort of, the appearance can help finding hands by the body pose. And the body pose can do the appearance. So if you think, for example, that, initially, I learned this particular hand in this particular appearance and pose, then I learn it by appearance. I learn it by the pose. But then if, for example, I keep the same pose but change completely the appearance of the hand, I still have the pose guiding me to the right location. And then I grab a new image and say, OK, this is still a hand, but a new appearance. Now that I have the new appearance, I can move the new appearance to a new location. So now I can recognize the hand by the appearance that I already know. But then, this is a new pose, so I say aha. So this is still a new body configuration that ends in a hand. And then I can change the appearance again. So you can see that by having enough images, I can use the appearance to learn various poses that end up in a hand using the common appearance and vice versa. If I know the pose, I can use the same pose and deduce from the different appearances of the same hand. And this becomes-- I will not go into the algorithms, but this becomes very powerful. And just by going through this iteration in which you start from a small subset of correct identification, but then you let these two schemes guide each other-- this kind of learning that one system guides the other-- we see, here, graphs of performance. Roughly speaking, the details are not that important. This is called a precision recall graph. But even without explaining the details of recall and precision, the higher graphs mean better performance of the system. And this is the initial performance of what you get if you just train it using hands grabbing objects. Actually, the accuracy of the system-- the system is doing a good job at recognizing hands, but only in a limited domain of hands touching objects. So other things, it does not recognize very well. So this is shown here, that it has high accuracy, but does not cover all the range of all possible hands that you can learn. And then without doing anything else, we just continue to watch movies. But you also integrate these two systems. Each one is supplying internal supervision to the other one. Everything grows, grows, grows. And after a while, after training with several hours of video, we get up to the green curve. And the red curve is sort of the absolute maximum you can get. This is using the best classifier we can get. And everything is completely supervised. So on every frame, in 10,000 frames or more, you tell the system where exactly the hand is. So this is what you can get with a completely supervised scheme. And this, the green, is what you can get with reasonable training-- I mean, seven hours of training. Infants get more training-- completely unsupervised. It just happens on its own. It's interesting, also, to think about-- if you think about infants-- and actually, I wanted to-- I was planning to ask you the question here let's see. What else could help infants do-- if you can think of other tricks in which infants somehow have to pick up, it's very important for them to pick up hands. What other signals, or tricks, or guidelines could help them pick up hands? And OK, since I sort of gave up own hands, you can think about babies sort of waving their hands. And babies do wave their hands a lot in the air. And you can think of a scheme in which the brain knows this. And you sort of wave hands. And then the images that are generated by these motive activities interpreted by the system, it already knows, grab this, and somehow, use them in order to build hand detectors. There is evidence that this is, I think, interesting. And we know that infants are interested in their own hands. But there are reasons to believe that this is not the case. Because for example, if you really try this, and you try to learn hands from waving your hands in this way, imitating what infants may see, a scheme that learns hands in this way is very bad at recognizing hands manipulating objects. If, after waving the hands, you test the system, and there is a hand coming in the image and touching, grabbing something, the difference in appearance in point of view between waving your own hands and watching somebody grabbing an object is so large that it does not allow to generalize well at this stage. And we know, from testing infants, that the first thing that they recognize well is actually other hands grabbing objects. So it's much more consistent with the idea that the guiding internal signal that helps them deal, in an unsupervised way, with this difficult task is the special event of the hand as the mover of objects. OK, I want to move from hands to-- this will be shorter, but I want to say something about gaze. Gaze is also, as I said, interesting. It starts at about three months of age. An infant has this capability. What happens at three months of age is that an infant may look at an adult-- at a caregiver, the mother, say, or look at another person-- and if the other person is looking at an object over there, then the infant will look at the other person, and then will follow the gaze, and will look at the object that the other person is looking at. So it's, first of all, the identification of the correct direction of gaze, but also then using it in order to get joint attention. And all of these things are known to be very important in early visual development. And psychologists, child psychologists, talk a lot about this mechanism of joint attention in which the parent, or the caregiver, and the child can get joint attention. And some people, some infants, do not have this mechanism of joint attention, being able to attend to the same event or object that the other person is attending to. And this has developmental consequences. So it's an important aspect of communication and learning. So understanding direction of gaze is very important. And here it's even perhaps more surprising and unexpected even compared to hands. Because gaze, in some sense, doesn't really exist objectively out there in the scene. It's not an object, a yellow object, that you see. It's some kind of an imaginary vector, if you want, pointing in a particular direction based on the face features. But it's very non-explicit. And you have, somehow, to observe it, and see it, and start extracting it from what you see, and all of this in a completely unsupervised manner. So what would it take for a system to be able to watch things, get no supervision, and after a while, extract correctly direction of gaze? Direction of gaze is actually depend on two types of sources of information, one of the direction of the head, and the other is the direction of the eyes in the orbit. And both of them are important. And you have to master both. There are more recent studies of this, and more accurate studies of this, but I like this reference. Because this is from a scientific paper on the relative effect of hand-- of head orientation and eyes' orientation in gaze perception. And it's a scientific paper in 1824. So this problem was studied with experiments, and the good judgment, and so on. The point here, by the way, is that these people look as if they are looking at different directions. But in fact, the eyes here are exactly the same. It's sort of cut and paste. It's literally the same eye region, and only the head is turning. And this is enough to cause us to perceive these two people as looking in two different directions. In terms of inference and how this learning comes about, the head comes about first. And initially, if the caregiver, as I said, is-- the head is pointing in a particular direction but the eyes are not, the infants will follow the head direction. Later on-- so this is at three months. Later on, they combine the information from the head and from the eyes. And the eyes are really subtle cues which we use very intuitively, very naturally. But although it's-- let me hide this for a minute. This person-- it's a bad image. It's blurred, especially those who sit in the back. But this person, is he looking, basically, roughly at you, or is he looking at the objects down there? Sorry? AUDIENCE: audience SHIMON ULLMAN: Yeah, basically at you, right? Now, if you look at the-- so it's from the eyes. And this is the-- these are the eyes. This is all the information. This is the right pixels, the same number of pixels that are in the image, and so on. So this is the information in the eyes. It's not a lot. And we use it effectively in order to decide where the person is-- we just look at it. And it's interesting that it's so small and inconspicuous in some objective terms. But for us, we know that the person is looking, roughly, at us. Now, we have some computer algorithms that do gaze. And gaze, again, it's not an easy problem. And people have worked quite a lot on detection, detecting direction of gaze. And all the schemes are highly supervised. And once it's highly supervised, you can do it. So by highly supervised, I mean that you give the system many, many images in which you give the image, but you also give the learning system-- together with the input image, you supply it with the direction of gaze. So this is the image, and this is the direction the person is looking at. And there are ways of getting this input information of the appearance of the face and the direction of gaze, to associate them, and then when you see a new face, to recover the direction of gaze. But it really depends on a large set of supervised images. So we need something to-- if you want to go along the same direction that happened before with getting hands correctly, we want-- instead of the internal supervision, we want-- some kind of a signal that somehow can tell the baby, can tell the system-- without any explicit supervision, provide some kind of an internal teaching signal that will tell it where is the direction of gaze. It's very close to the hand and the mover using the following notion, that if I pick up an object, which was very close to the mover event, once I picked up an object, I can do whatever I want with it. I don't have to look at it. I can manipulate it and so on. But if it's placed somewhere and I want to pick it up, nobody picks up objects like this. I mean, when you look at objects, when you pick them up, at the moment of making the contact to pick them up, you look at them. And in fact, that's a spontaneous behavior which we checked psycho-physically. You just tell people, pick objects. You don't tell them what you are trying to do. And they're invariably always-- at the instant of grabbing the object, making the contact, they look at it. So if we have, already, a mover detector, or sort of a hand detector, or a mover detector, that hands that are touching objects and causing them to move, all you have to do-- whenever you take an event like this, it's not only useful for hands. But once a hand is touching the object, this kind of mover event, you can freeze your video, you can take the frame. And you can know with high precision that you can-- now the direction of gaze is directed toward the object. So we asked people to manipulate objects on the table and so on. And what we did is we ran our previous mover detector. And let me skip the movie. But whenever this kind of a detection of a hand touching an object, making initial contact with an object happened, we froze the image. Unfortunately, this is not a very good slide, so it may be difficult to see. Maybe you can see here. So we simply drew a vector pointing from the face in the direction of the detected grabbing event. And we assumed-- we don't know-- that's an implicit, internal, imaginary supervision. Nobody checked it. But we grabbed the image, and we drew the vector to the contact point. So now you have a system-- on the one hand, we have face images at the point where we took the contacts. So here is a face. And this is a descriptor, some way of describing the appearance of the face based on local gradients. Sorry? AUDIENCE: How do you find the face? SHIMON ULLMAN: You assume a face. The face detector, I just left-- AUDIENCE: [INAUDIBLE] SHIMON ULLMAN: Right. Right. Faces come even before-- I didn't talk about faces, but faces come even beforehand. As I mentioned, the first thing that infants look at is faces. And this is even before the three months that they look at hands. This is to-- the current theory is that faces, you're born with a primitive initial face template. There are some discussions where the face template is. There is some evidence that it may not be in the cortex. It may be in the amygdala. But there is some evidence for this face. For manipulation of these patterns that infants look at, it's a very simple template, basically the two eyes, or something round, with two dark blobs. And this makes them fixate more on faces in a very similar way to the handling the hands. You just-- if you do this, from time to time, you will end up focusing not on a face, but on some random texture that has these two blobs or something. But if you really run it, then you will get lots and lots of face images. And then you'll develop a more refined face detector. So babies, from day one by the way, the way we think that-- the way people think it's innate is you can work-- it is done-- experiments have been done with the first day when babies were born, the day one. They keep their eyes closed most of the time. But when they are open, they fixate. You have to make big stimuli, because it's like close-up faces, because the acuity is still not fully developed. But you can test what they're fixating on. And they fixate specifically on faces. And once there is a face, they fixate on it. And the face can move, and they will even track it. So this is day one. So face seems to be innate in a stronger sense. In the case of the hand, for example, as I said, you cannot even imagine building a innate hand detector because of all this variability in appearance. For the face, it seems that there is an initial face detector which gets elaborated. So we assume that there is some kind of-- in these images, we assume that when we grab an event of contact like this, the face is known, the location of the face, the location of the contact is known. And you can draw a vector from the first to the second. And this is the direction of gaze. And when, now, you see a new image in which there is no contact, you just have the face, and you have to decide what is the direction of gaze, you look at similar faces that you have stored in your memory. And from this stored face in memory, for this, you already know from the learning phase what is associated direction of gaze. And you retrieve it. And this is the kind of things that you do. What we see here with the yellow arrows are collected images, which, again, the direction of gaze, the supervised direction, has been collected, or was collected, automatically, by just identifying the direction to the contact point. These are some examples. And what this shows is just doing some psychophysics and comparing what this algorithm-- which is sort of this infant-related algorithm which just has no supervision, looking images for hands touching things, collecting direction of gaze, developing a gaze detector. So the red and the green, one is the model, the other one is human judgment on a similar situation. And you get good agreement. I mean, it's not perfect. It's not the state of the art, but it's close to state of the art. And this is just training with some videos. I mean, this certainly does at least as well as infants. And it keeps developing, getting from here to a better and better gaze detector with reduced error. Well, the error is pretty small here too. But you can improve it. That's already-- that's more of standard additional training. But making the first jump of being able to deal with this nonexistent gaze, collecting a lot of data without any supervision, which is quite accurate, about where the gaze is and so on, this is supplied by, again, this internal teaching signal that can come instead of any external supervision and make it unnecessary. And you can do it without the outside supervision. It also has, I think, some-- the beginning of the more cognitive correct association, like the hand is associated with moving object, direction of gaze and going to where the-- following it to see what is the object at the other end and so on, this is-- gaze is associated with attention of people, what they are interested in at the moment. So it's not just the fact that you connect the face with the target object and so on. It's a good way of creating internal supervision. But it also starts to, I think, create the right association that hand is associated with manipulation and goals of manipulating objects. And gaze is associated with attention, and what we are paying attention to, and so on. So you can see that you start to have-- based on this, if you have an image like this, and you can detect hands, and you know, what-- there's a scheme that does it. You know about hands. You know about direction of gaze. You know about-- I didn't talk about it, but you also follow which objects move around. And you know which objects are movable and which objects are not movable-- so a very simple scheme that follows the chains of processing that I described. So it already starts to know-- you know, it's not quite having this full representation in itself. But it's thought quite along the way that the two agents here. The two agents are manipulating objects. And the one of the left is actually interested in the object that the other one is holding. So you have all this, the building blocks to start build-- to start having an internal description along this line following the chain of processing that I mentioned. And by the way, this internal training that one thing can train another, if you want, simple mover can train the hand. Mover and a hand together can train a gaze detector. It turns out that gaze is important in learning language, in disambiguating nouns and verbs when you learn language, when you acquire your first language. So this is from verb learning for a particular experiment. But let me ignore that. A simple example would be acquiring a noun that I say. I say, suddenly, oh, look at my new blicket. And people have done experiments like this. And I can say, look at my blicket, looking at an object on the right side or looking at another object on the left side, saying exactly the same expression. And people have shown that infants exposed to this kind of situation, they automatically associate the term, the noun "blicket," with the object that has been attended to. Namely, the gaze was used in order to disambiguate the reference. So you can see a nice-- starting with very low-level internal guiding signals of, say, moving pixels that can tell you about hands and about direction of gaze-- and then direction of gaze helps you to disambiguate the reference of words-- so these kinds of trajectories of internal supervision that can help you learn to deal with the work. This is, to me, a part of a larger project, which we called the digital baby, in which we-- it's an extension of this. We really want to understand, what are all these various innate capacities that we are born with cognitively? And we mention, here, a number of suggested ones-- the mover, how the mover can train a gaze, and the core training of two systems. And some of the things we think that are happening innately before we begin to learn. And then we would like to be able to watch lots and lots of sensory input, which could be visual. It can be, in general, non-visual. And from this, to generate, what will happen is the automatic generation and lots of understanding of the world, concepts like hands and intention, direction of looking, and eventually, nouns, and verbs, and so on-- so how we'll be able to do it. Know that it's very different from the less structured direction of deep networks, which are interesting and are doing wonderful things. I think that they're a very useful tool. But I think that they are not the answer to the digital baby. They do not have the capacity to learn interesting concepts in an unsupervised way. They do not distinguish. They go, as I showed you at the very beginning, with the cats, and the upper body, and so on. They go only for the salient things. Gaze is not a salient thing. I mean, we have internal signals that allow us to zoom in on meaningful things. Even if they are not very salient objectively in the statistical sense, there is something inside us that is tuned to it. We are born with it. And it guides us towards extracting this meaningful information, even if it's not all that salient. So all of these things are missing from the unstructured net-- or the networks which do not have all of this pre-concept and internal guidance. And I don't think that they could provide a good model for cognitive learning in this sense of the digital baby. Although I can see a very useful role for them, for example, as just-- in answer to Doreen's question, that if you want to then get-- from all the data and the internal supervision that you provided, you want to get an accurate gaze detector, then training, using supervision training in appropriate deep networks can be a very good way to go. I wanted to also show you, this is not directly related, but it's something impressive about the use of hands in order to understand the world, just to show you how smart infants are. I talked more about detecting the hands. It was more the visual aspect of, here is an image, show me the hand. But how they use it-- and this is at the age of about one year-- maybe 13 months, but one year of age. Here's the experiment. I think it's a really nice experiment. This experiment was with a experimenter. This is the experimenter, one image. What happened in the experiment is that there was a sort of a lamp that you can turn the lamp on by pressing it from above. It sort of has this dome shape. That's the whites in here. You press it down, and it turns on. It shines blue light, and it's very nice. And babies like it. And they smile at it, and they jiggle, and they like this turning of the bright light. And during the experiment, what happens is that the infant is sitting on its parent's lap. And the experimenter is on the other side, that experimenter. And she turns on the light. But she turns on the light-- instead of pressing it, as you'd expect, with her hand, she's pressing on it with her forehead. She leans forward, and she presses the lamp, this dome, and the light comes on. And then, these are babies that can already manipulate objects. So after they see it three or four times, and they are happy seeing the light coming on, they are handed the lamp and asked to turn it on on their own. And here is the clever manipulation. For half the babies, the experimenter had her hands concealed. She didn't have her hands here, you see? No hands are under this poncho. Here it's the same, very similar thing, but the hands are visible. Now, it turns out that the babies-- or the in-- these are not babies anymore. These are young infants-- some of them, when they were handed the lamp, they did exactly what the experimenter did. They bent over, and pressed the lamp with their forehead, and turned it on. And other children, instead of-- although that's what they saw, when they got the lamp over to their side, they turned it on by pressing it with their hand, unlike what they saw the experimenter do. Any prediction on your side what happened-- you see these two situations-- when which babies-- I mean, in this case or in this case, in which case do you think they actually did it with their hands rather than using their forehead? Any guess? Yeah? AUDIENCE: hands in A and no hands in B SHIMON ULLMAN: That's right. And what's your reasoning? AUDIENCE: [INAUDIBLE]. SHIMON ULLMAN: But you think about, you know, baby, if you saw a baby, infant, young one-year-old just moving seemingly quasi-randomly and so on, something like that went on in their head, that here, she did it with her forehead. She would have used her hands, but she couldn't, because they were concealed, and she couldn't use them. So she used her forehead, but that's not the right way to do it. I can do it differently and so on. That's sort of-- they don't say it explicitly. They don't have language. But that's the kind of reasoning that went on. And indeed, a much larger proportion-- so this is the proportion of using their hands where the green, I think, was the hand occupied. Or when the hands of the experimenter is free, you see that there is a big difference between the two groups. So they notice the hands. They ran through some kind of inference and reasoning. What hands are useful? Why I should do? Should I do it in the same way because that's what other people are doing? Should I do it differently? So I find it impressive. So some general comment is-- general thoughts on learning and the combination of learning and innate structures, that there is a big sort of argument in the field, has been going on since philosophers in ancient times, whether human cognition is learned. This is nativism against empiricism, where nativism proposed that things are basically-- we are born with what is needed in order to deal with the world. And empiricism in the extreme form is that we are born with a blank slate and just a big learning machine, maybe like a deep network. And we learn everything from the contingencies in the world. So this is the empiricism versus nativism. In these examples, in an interesting way, I think, that complex concepts were neither learned on their own nor innate-- so for example, we didn't have an innate hand detector, but also, it couldn't emerge in a purely empiricist way. But we had enough structure inside that would not be the final solution, but would be the right guidance, or the right infrastructure, to make learning possible. And this is not just a very generic learner. But in this case, the learner was informed by-- you know, was looking for some mover events or things like it. So it's not the hands, it was the movers. And this guides the system without supervision, not only making supervision unnecessary, but also focusing the learner on meaningful representations, not necessarily just things which are the first things that jump at you statistically from the visual input. So there are these kinds of learning trajectories like the mover, hand, gaze, and reference in language, of sort of natural trajectories in which one thing leads to another and help us acquire things which would be very difficult to extract otherwise. As I mentioned at the beginning, I think that there are some interesting possibilities for AI, as I said, to build intelligent machines by not thinking about the final intelligence system, but thinking about baby system with the right internal capacities which will make it able to then learn. So the use of learning is sort of-- we all follow it. But the point is, probably just a big learning machine is not enough. It's really the combination of, we have to understand the kind of internal structures that allow babies to efficiently extract information from the world. If we manage to put something like this into a baby system and let it interact with the world, then we have a much higher chance of starting to develop really intelligent systems. It's interesting, by the way, that in the regional paper by Turing, when he discusses the Turing test and how to build-- can machines think, he discusses the issue about building intelligent machines somewhere in the future. And he says that his hunch is that the really good way of building, eventually, intelligent computers, intelligent machines would be to build a baby computer, a digital baby, and let it learn rather than thinking about the final one.
MIT_RES9003_Brains_Minds_and_Machines_Summer_Course_Summer_2015	Lecture_17_Hippocampus_Memory_Sleep_Part_2.txt	The following content is provided under a Creative Commons license. Your support will help MIT OpenCourseWare continue to offer high quality educational resources for free. To make a donation or view additional materials from hundreds of MIT courses, visit MIT OpenCourseWare at ocw.mit.edu. MATT WILSON: Now, why are we interested in sleep? So, we kind of think about this as two modes, online mode and offline mode. In the online mode, you're taking information in. In the offline mode, you're going back and evaluating information. What's the purpose of evaluating information that you've taken in? Well, if we think about the general problem that's being solved, the problem of intelligence. The problem of intelligence is trying to understand and infer these sort of deep generalizable relationships rules. You're trying to extract rules from instances. And you'd like those rules to be as generally applicable as possible. And in order to do that, presumably, one has to go back and evaluate the many individual instances to try to extract some kind of statistical regularity, and then perhaps evaluate models that you have constructed in terms of their consistency with individual instances that you've already collected. Or perhaps future instances that you haven't yet. So you have the raw material, you build a little model, and then you continually test that model against new information that comes in. And the question is, when can you do that? Now, you could do that when you're out and about in the world. As you saw and read in that paper, when animals are sitting quietly, they very quickly can switch into this kind of offline mode that looks a lot like sleep. In fact, electrophysiologically, sleep and quiet wakefulness in the hippocampus are nearly indistinguishable. It's the same kind of offline mode. It says, OK, when the hippocampus is not being used to take new information in, I quickly switch into this offline evaluation mode. But during sleep, I no longer have the constraint of having to direct behavior. Behavior's shut off, inputs are shut off, and now I can switch into this purely internal introspective mode. So what goes on during sleep. In this simple experiment, we look at the activity when the animal is performing behavioral tasks. And then we examine activity during sleep, both before and after, and ask, is there anything about behavioral experience that changes activity during sleep? And what we found was that if you look at activity during behavior in which you think about these spatial sequences being expressed, and you look during sleep afterward, you find that the spatial sequences are expressed again. So the hippocampus replays the firing of these cell sequences. But it replays the firing of these cell sequences at a time scale that appears to be compressed relative to the behavioral timescale. So these are eight place cells. Animals were walking from left to right. The ticks indicate that's the location of the peak. So these place cells will fire one through eight as the animal moves along the track over about five seconds. That's how long it takes the animal to walk from left to right. The same sequence of 1, 2, 3, 4, 5, 6, 7, 8 gets replayed during sleep, but now over about 150 milliseconds. Same sequence, so you're preserving time order, not absolute time. And that when you ask, when these sequences are expressed, what's going on in the local field potential in these oscillations? And this is where you find these sharp wave ripple events. This is this ripple-like event that I was describing to you. So these sharp wave ripples are when these apparently compressed sequences are being expressed. I could point out that a simple model for these sharp wave ripple reactivated sequences is the same model that I like to use to explain phase precession during the fade oscillation. That is, I give it an input, I sweep inhibition from high to low such that cells that are getting the strongest input fire earlier. And this will reactivate a sequence. So the difference between a theta sequence and a reactivated sequence is really just, where's the input coming from. If the input's coming from what I'm actually experiencing right now, now it's a theta sequence. And as I move, the input changes systematically. And so, again, it looks like it's encoding information about space. If I'm now offline, and this information is being delivered to the hippocampus from some other source, you can apply exactly the same operation, get the same kind of sequence. It's just now it's on information that is not tied to your immediate context or location. That's the only difference. Where is the input coming from? Now further, if you think about this model, you can imagine the depth of disinhibition could actually affect the length of the sequence. But that's just sort of a mechanistic thing. And if you actually look at the mechanisms that regulate sharp wave ripples in the theta oscillation, it basically comes from the same structure. It's a structure that regulates acetylcholine, and it's a neuromodulator that's associated with attention and memory, and the structure called the medial septum. So the medial septum provides the drive, the cholinergic drive, of the hippocampus. Damage to the medial septum, loss of cholinergic tone, was one of the dominant models of neurodegenerative cognitive and memory loss in Alzheimer's disease. So what you find is one of the earliest indications of neurodegenerative damage in Alzheimer's is the loss of cholinergic tone. And the systems that begin to break down are the systems that actually fall along this limbic pathway starting with hippocampus-entorhinal cortex. So it's as though the cholinergic system that regulates the expression of this oscillation in the hippocampus, when it breaks down, it leads to general memory loss and disruption. And it turns out the medial septum is also involved in regulating the expression of these sharp wave ripples. So, same system, different modes. One quite active, online. One inactive, offline. Same kind of modulation of excitability through inhibition. But that's the idea. Simple model, sweep inhibition, that gives you the sequences. And then if you can control the input, you control the input to control the content, you control the inhibition to control the structure of the timing. Those are the two things. So the question is, how could you control the input? Well, you're kind of thinking about what is the input into the hippocampus under these two conditions. And as I mentioned, hippocampus, you've got the entorhinal cortex. Entorhinal cortex gets information across the brain. It's sort of these sensory association areas, visual cortex, auditory cortex. So all this information about the world converging on the hippocampus and then getting modulated. And so let's look in, for instance, the visual cortex. So if we simultaneous record visual cortex and the hippocampus, we could see how these two structures communicate. And when we do this in a simple task-- this is like a little figure eight task-- and I won't go into a lot of the details. But one interesting thing that came out from doing this experiment, recording the visual cortex and the hippocampus, is that recording in the visual cortex when an animal is moving in space, what you find is cells in the visual cortex have spatial-like receptive fields. Similar to the hippocampus, though not as spatially tuned. But here, for instance, are eight visual cortical cells. And you can see that they fire in a sequence. They'll have different spatial receptive fields. This shows where this one visual cortical cell likes to fire. Different visual cortical cells will fire at different locations. So when animals are moving in space, you actually see sequential activation of these visual cortical responses. And so if you have visual perceptual sequences in hippocampal spatial sequences, one question is, how do those sequences relate offline when the animal, for instance, during sleep. So you have sequences during sleep in the hippocampus, sequences during sleep in the visual cortex. And it turns out that those two things are actually correlated. So when hippocampus plays out a spatial sequence, the visual cortex plays out a visual sequence that corresponds to the visual responses at those locations. So hippocampus plays out a sequence of where it was, visual cortex expresses the visual stimuli that were present along that sequence. One thing about these sequences when we're looking at hippocampal neocortical interactions is that the sequences are now at a much longer time scale. And in fact this time scale, which here is on the order of about half a second to a second, corresponds to another oscillation that is characteristic of sleep known as the slow oscillation. So when you go to sleep, brain rhythm, the oscillations, the dominant frequencies start to slow down. And you'll get this oscillation in this one hertz or so frequency range. If you record activity from a bunch of cells, it looks like this. This is the activity of a whole bunch of cells in the visual cortex in the hippocampus. And you see that the activity is flipping between lots of cells active, no cells active. These are the so-called up and down states of cells during sleep. That every half a second to a second or so many cells will become active, then they'll all be shut off, then they'll become active again. So you are flipping between these up and down-like states. And if you look at these up and down states in the visual cortex, and you look at similar activity in the hippocampus, you see these up and down-like states in the hippocampus as well. But you'll notice that the up state in the neocortex leads up state in the hippocampus. So neocortex first, then hippocampus, neocortex, hippocampus. Neocortex seems to lead. So again, this question of, who's providing the input? That simple model where I'm sweeping inhibition during the theta, during active behavior, that information is coming in from perception, is what I'm saying. During sleep, the question is, where is that information coming from? Well, this would say, this is where it's coming from. It's coming from these cortical areas. Let's say sensory, visual cortical areas turn on. They provide the information to the hippocampus. Now the hippocampus turns on, and the hippocampus responding to input coming in from the sensory cortex. Now, if that's the case, could we manipulate hippocampal activity by manipulating the sensory cortex? And so this experiment, simple experiment, answer was yes. In this case we use the auditory system. One of the reasons to use the auditory system is that, unlike the visual system, the auditory system remains in a state of persistent vigilance, even during sleep. Auditory cortical responses, even when the animal goes to sleep, measure auditory cortical responses same as when it's awake. So essentially the auditory cortex stays vigilant during sleep, which has clear evolutionary value. Animal's asleep, it's trying to minimize arousal by shutting off visual input, which may be of limited value anyway, given the circadian nocturnal nature of visual stimuli. So if you can't see anything, you might as well actually close your eyes. But you can still hear things. So you and other animals are still listening to pick up threats that might require that they actually wake up. So anyway, taking advantage of that. Auditory system on, so train an animal on task where it learns to associate auditory cues with locations. Right sound means go over here to get food, left sound means go over there to get food. Very simple task. Then animal goes to sleep, and you just continue to play the sounds. So it's learned something. And now you try to bias cortical activity during sleep and ask, what does that do? And so this is the idea. When the animal is actually running, we can decode hippocampal activity. So we can tell, oh, here's the hippocampal pattern that corresponds to the left side or the right side. And in this case, when the animal's performing the task, play the right-hand sound, animal goes to the right-hand side, see this in the hippocampal response. Play the left-hand sound, animal goes to the left-hand side. Again, the left-hand place fields when the animal's on the left, right-hand place fields on the right. So this is what the experiment and the behavior looks like from the standpoint of the hippocampus. Very clear, right? Right sound, right hippocampus, right place cells. Left sound, left place cells. But now the animal's going to go to sleep, and we're going to do the same thing. Continue to play the right sounds and the left sounds. It's just that now the animal is not actually moving on the track. And so we ask, does that-- can we bias the reactivated sequences that you get? So there's the same thing now. Animal's awake, but now it's going to go to sleep, and we just keep playing the sounds. The little tics there indicate the delivery of sounds every 10 seconds or so. So we play a sound, and now we look at the response. The difference here is that the animal's not running, it's not behaving. Sound, response. Now we'll decode the activity. So here, for instance, we take a little short window here, about half a second. And what you see is this is this up-like state. Multi-unit activity, lots of cells firing. Decode activity. Now ask when you play the left sound, what does activity decode to? And here you see it's going to decode into the left-hand side. So that's the basic hypothesis. Play the left sound, you get the replay of the left side of the track. You play the right sound, you get reactivation of the right side of the track. And that's what you get. So when you play the left-hand sound, left sound bias activates place cells on the left side. Right side bias biases place cells on the right. And you can do the same kind of psychophysical experiments which has been done in humans, either with different sensor modalities-- it's been done in olfaction. It's also been done in audition. So the equivalent experiment, using auditory cueing in humans, where you have people learn the simple task. And this was done in Ken Paller's lab where they do the simple spatial matching game. It's like, you know, where's the cat card, where's the teapot card? The variant here is when they flipped over the cat card, they would play the cat sound, a meow. When they flip over the teapot sound, they play a little associated auditory cue, teapot whistle. And then people go to sleep. And during sleep, they would play either the cat sound or the teapot sound. And they found that when they would wake up and now they do the task, if they played the cat sound during sleep, they were better at remembering the location of the cats than the whistle. So this says not only does sleep actually contribute to memory, but it's selective. And not only is it selective, but it can be influenced. It can be biased. You can direct the nature memory processing. And then our experiments suggest that, well, one of the consequences of this kind of sleep manipulation would be to bias the memory reactivation in the hippocampus. And so the idea is simple. That is that cortex biases the state that the hippocampus gets. And then the hippocampus takes that state, sweeps inhibition, replays a sequence, and that sequence then gets played back to the cortex. So the cortex, it sort of knows-- it has these sort of discrete states that doesn't necessarily know what the causal correlations might be. What happens next? It doesn't necessarily know what happens next. It knows what happened. Hippocampus knows what happened next. It has lots of instances of that though. Well, I saw a red light, what happens next? You say, oh, you know, I saw red light, and all the cars stopped. OK, that's great. If I'm just the cortex, that's what I learned. All the car stop. If I'm the hippocampus, I'm a little bit smarter than that. I say, you know, last Tuesday I was there, there's a red light, and all the cars, except for these cyclists. Man, they didn't stop, they just kept on going. So wait a minute, there's a rule. Red light, cars stop, bicycles don't stop. So you have to refine what appear to be simple rules. Often that's actually used in an example. How do you use the prefrontal cortex? Oh, red light means stop, green light means go. That's great, except in the real world, that rule is too simple. It has to be refined based upon your particular experience. In fact, if you're really sophisticated, red light means cars, except for cabbies, will stop, right? If I see a cabbie, guaranteed that guy's not going to stop, he's going to accelerate, right? And that's the kind of information-- that's what the hippocampus has. And so you imagine that's what's going on. Neocortex, in each one of these slow oscillations, is saying red light. And the hippocampus says, oh, yeah, OK, cars stop. Red light again. Well, there was that bicycle thing, bicycles continue to go. Red light. Well, that was the cabbie incident. So now you have all these sort of causal sequences that are being expressed back to the neocortex, presumably in order to establish this more comprehensive, consolidated model of real world traffic lights, rather than the cartoon traffic lights. And so that would be the idea of what's going on during sleep. Now, during quiet wakefulness you can think of the same sort of idea. And that is that you're kind of evaluating multiple casual contingencies. Each of which might be expressible as a simple rule, but might also be experienced as distinct variations of that rule. And the idea is, OK, do we use the rule, or do we use the exceptions, or how do we actually refine the rule based upon these different instances or exceptions? And so you can think about that as refining the rule, that's like the learning side. Applying the rule, that's the memory or decision making side. So you can think about, during quiet wakefulness, this reactivation being used in the service of actual learning or in decision making. And so again, you read the paper, and one of the things that we had discovered interestingly about reactivation during quiet wakefulness was that when animals stop after running on a track, they do reactivate sequences. This is raw data, animal running from left to right and then stopping for a long period of time. And you can see activity. There are these bursts of activity. You blow these things up, these are these sharp wave ripples. They last about half a second or so. So you see a burst of activity, you see these place cell sequences. In this case the sequence actually runs in time reversed fashion. So in this case-- time reversed fashion-- this doesn't seem like planning or decision making. It may be evaluation of temporal correlations that might be relevant to behavior and learning. But what would a reverse sequence actually have to do with spatial learning and memory? The insight into how this might be used came from computational models. In fact, computational models that the post-doc in this case, Dave Foster, who made this discovery, had used in his doctoral work. Where he was actually building models, reinforcement learning models of spatial navigation. And one of the problems in reinforcement learning is that reinforcement often comes after you've actually carried out the steps that lead up to it. In other words, you walk from left to right, you get rewarded when you get here. What you want to know is not just, this is where the reward is. What you really want to know is, what were the things that actually lead up to that? In other words, I want to take credit, reward value, and I want to spread it backward in time to place value on the things that predict or lead up to reward. The so-called temporal credit assignment problem. How do I give credit to things that actually lead up to or predict reward? And thinking about how you might solve the temporal credit assignment problem, this reverse reactivation actually has the capacity to solve that problem in one step. If you imagine when the animal gets to the end, gets some reward, and if now at that point I pair the delivery of reward signal, which I will indicate here as-- this is a cartoon suggesting this is dopamine, a reward signal. And I'm going to pair that with the reverse sequence. And now you can think of the association. This is my current location. This is way back where I started from. Current location gets paired with strong reward. Remote location gets paired with low reward. So this association will essentially solve, will convolve this discrete reward impulse function, will turn it into this continuous graded monotonic reward gradient function. Translating this into this essentially in one step. So the thinking is, hey, animals are actually using this to learn, and they're using this to solve the temporal credit assignment problem in a way that would be important for general reinforcement learning. Now, I won't go into-- so, we actually did this experiment recorded from the reward area, the VTA, and the hippocampus. And, indeed, during these reactivation events, you see the pairing of this reward signaling. And not only do you see the pairing of the reward signaling, but you find pairing of the reward signaling in which the precise firing of reward signals map onto the delivery of rewards at goal locations. So it's not just certain sequences are good, others are bad. It's that there are certain locations along the sequence that have differential reward value. So there's a mapping of relative reward to location. And that if you look at this while animals are actually performing a task you can see biases in these sequences that correspond to planning. So both of these elements of sequence reactivation-- this was some work by Dave Foster where he looked at an animal that has to just forage and find a location space. When animals are searching for a location, and you look at these reactivated sequences, it turns out that reactivated sequences tend to be directed toward the locations where the animal thinks the goal might be. So it's sort of thinking about things that would lead to reward. So both sides of this kind of sequential computation seem to be expressed in the hippocampus. Co-expression with reward during evaluation or learning. The expression during reward directed behavior in the service of planning and decision making. And then finally, this is the paper that you had read. It's just looking at the structure of these reward sequences on long tracks. And the most salient point that came out of this paper is that the phenomena of sharp wave ripple sequential activation that correspond to these theta sequences, when animals are sitting quietly over the course of these longer up state like events, takes the form of bursts of sharp wave ripples, or sequences of these short sequences. So there's this compositional notion that you form long sequences out of little short sequences. And the other question is, how do you actually put these short sequences together, and why would you have a representation that has this kind of compositional structure? It's the Lego block kind of idea, where I can put together sequences from these more elemental sequential units. I won't show the movie, but if I actually zoom in on one of those sequences in that particular instance where the animals stop, you can see there's a longer reactivated sequence that actually takes the form of four shorter sequences. That's how long sequences are being evaluated. And one interesting thing about this was that if you look at these different sequences of different length, shorter sequences, longer sequences, they all seem to have a fixed velocity. And that is, there's this kind of a fixed-- what I've heard is the speed of thought. So there is some fixed constraint that evaluating further out in the future simply takes more time. And so that suggests that there's an interesting constraint in capacity for extended evaluation that comes in the form of these oscillatory modes of the slow oscillation. Longer, or slower, frequencies give you, essentially, more time, longer sequences. And that, potentially, if you want to go even longer, you might actually even link sequences across subsequent cycles. Just as you could do across successive sharp wave ripples, you might be able to do that across successive slow waves. So the idea that you have these oscillations that you can link or couple sequences across different cycles of these oscillations, that you have oscillations at different frequencies, suggests that this is the mechanism. It's the compositional mechanism for sequences that are nested within oscillations. By combining these oscillations you can construct these sequences in ways that presumably contribute to cognition. So we now kind of see how things are structured, and now the question is, can we actually manipulate it? And so that's really the challenge. Seeing the compositional structure, having demonstrated potential access, the ability to bias content. To coord, to bias the structure of when things occur, what actually occurs. Now I think the capacity exists to essentially engineer the sequences themselves. And that is that, during the sleep or quiet wakeful states, get animals to think about things that they have not actually experienced, and test the hypothesis that this is what animals are using to construct these models, to essentially tinker with the blocks of memory and cognition. And then finally, just pointing out, it's interesting. These ripple sequences, same space and time scale as theta sequences. So again, it's suggesting this is the fundamental unit. It's not unique to the hippocampus. Essentially any brain system could express this kind of structure using that simple model. Where you see ramps and you see oscillations, you're going to get sequences. And this capacity is something that probably is broadly expressed and enjoyed by different brain areas. So, there you go.
MIT_RES9003_Brains_Minds_and_Machines_Summer_Course_Summer_2015	Lecture_84_Stefanie_Tellex_HumanRobot_Collaboration.txt	The following content is provided under a Creative Commons license. Your support will help MIT OpenCourseWare continue to offer high-quality educational resources for free. To make a donation or view additional materials from hundreds of MIT courses, visit MIT OpenCourseWare at ocw.mit.edu. STEFANIE TELLEX: So today I'm going to talk about human robot collaboration. How can we make robots that can work together with people just as if they were another person and try to achieve this kind of fluid dynamic that people have when they work together? These are my human collaborators. This work is done by a lot of collaborative students and postdocs. So we're really in an exciting time in robotics because robots are becoming more and more capable and they're able to operate in one structured environment. Russ gave a great talk about Atlas doing things like driving a car and opening doors. This is another robot that I've worked with. A robotic forklift that can drive around autonomously in warehouse environments. It can detect where pallets are, track people, pick things up, put things down. And it's designed to do this in collaboration with people who also share the environment. There's robots that can assemble IKEA furniture. This was Ross Knepper and Daniela Rus at MIT that I worked with to do this. So they made this team of robots that can autonomously assemble tables and chairs that are produced by IKEA. And what would be nice is if people can work with these robots. Sometimes they encounter failures, and the person might be able to intervene in a way that enables the robot to recover from failure. And kind of the dream is robots that operate in household environments. So this is my son when he was about nine months old when we shot this picture with a PR2. You'd really like to imagine a robot-- Rosie the robot from the Jetsons that lives in your house with you and helps you in all kinds of ways. Anything from doing the laundry to cleaning up your room, emptying the dishwasher, helping you cook. And this could have applications for people in all aspects of life. Elders, people who are disabled, or even people who are really busy and don't feel like doing all the chores in their house. So the aim of my research program is to enable humans and robots to collaborate together on complex tasks. And I'm going to talk about the three big problems that I think we need to solve to make this happen. So the first problem is that you need to be able to have a robot that can robustly perform actions in real-world environments. And we're seeing more and more progress in this area, but the house is kind of like this grand challenge. And John was talking about kind of all these edge cases. So I'm going to talk about an approach that we're taking to try to increase the robustness and also the diversity of actions that a robot can take in a real-world environment by taking an instance-based approach. Next, you need robots that can carry out complex sequences of actions. So they need to be able to plan in really, really large combinatorial state-action spaces. There might be hundreds or thousands of objects in a home that a robot might need to manipulate and depending on whether the person is doing laundry or they cooking broccoli or are they making dessert. The set of objects that are relevant that are useful that the robot needs to worry about in order to help the person is wildly different. So we need new algorithms for planning in this really large state- action space. And finally, the robot needs to be able to figure out what people want in the first place. So people communicate using language, gesture, but also just by walking around the environment and doing things that you can infer something about what their intentions are. And critically, when people communicate with other people, it's not an open loop kind of communication. It's not like you send a message and then close your eyes and hope for the best. People, when you're talking with other people, engage in a closed loop dialogue. There's feedback going on in both directions that acts to detect and reduce errors in the communication. And this is a critical thing for robots to exploit because robots have a lot more problems than people do in terms of perceiving the environment and acting in the environment. So it's really important that we establish some kind of feedback loop between the human and the robot so that the robot can infer what the person wants and carry out helpful actions. So the three parts of the talk are going to be about each of these three things. So this is my dad's pantry in a home. And it's kind of like John's pictures of the Google car. Most robots can't pick up most objects most of the time. It's really hard to imagine a robot doing anything with a scene like this one. There was just the Amazon picking challenge and the team that won used a vacuum cleaner, not a gripper to pick up the objects. They literally sucked the things up in the gripper and then turned the vacuum cleaner off to put things down. And the Amazon challenge had much, much sparser stuff on the shelves. We'd really like to be able to do things like this. And what we're doing now I'm really going to focus on a sub-problem, which is object delivery. So from my perspective, I think a really important sort of baseline capability for a manipulator robot is to be able to pick something up and move it somewhere else. We'd obviously love a lot more things. We're also talking about buttoning shirts in the car. And you can go on with all the things you might want your robot to do. But at least, we'd like to be able to do pick and place. Pick it up and put it down. So maybe you are in a factory delivering tools, or maybe you're in the kitchen delivering stuff like ingredients or cooking utensils. So to do pick and place in response to natural language commands-- so let's say, hand me the knife or something-- you need to know a few things about the object. First of all, you need to be able to know what it is. If they said, hand me the ruler, you need to be able to know whether or not this object is a ruler. So some kind of label that can hook up to some kind of language model. Second, you have to know where the object is in the world because you're going to actually move your grippers and your object and yourself through 3D space in order to find that object. So here I'm going to highlight the pixels of the object. But you have to register those pixels into some kind of coordinate system that lets you move your gripper over to that object. And then third, you have to know where on that object are you going to put your gripper. So in the case of this ruler, it's pretty heavy, and it's this funny shape that doesn't have very good friction. So for our robot, the best place to pick it up is in the middle of the object. And there might be more than one good place, and it might depend on the gripper. And different objects might have complex things going on that change where the right place is to pick it up. So conventional approaches to this problem fall into two general categories. The first category, the first high-level approach is what I'm going to call category-based grasping. This is the dream. So the dream is that you walk up to your robot, you hand it an object that it's never seen before, and the robot infers all three of those things, what it is, where it is, and where to put the gripper. And there's a line of work that does this. So this is one paper from Ashutosh Saxena, and there's a bunch of others. The problem is that it doesn't work well enough. We are not at the accuracy rates that sort of John was alluding to that we need for driving. So in Ashutosh's paper, I think they got 70% or 80% pick success rate on their particular test set at doing category-based grasping. And I think that you're going to have to expect that to fall if you actually give it a wider array of objects in the home. And even if it doesn't fall, 80% means it's dropping things 20% of the time, and that's not so good. The second approach is instance-based grasping. So I was talking to Eric Sudderth in my department who does machine learning, he said, instant recognition is a solved problem in computer vision. So instant recognition is I give you a training set of the slide flipper, lots of images of it, and then your job given a new picture is to draw a little box around the slide flipper. This is considered a solved problem in computer vision. There is a data set and a corpus, and the performance maxed out, and people have stopped working on it. And a lot of the work in robotics uses this kind of approach. We were talking about you have some kind of geometric model. That's the instance-based model. These models can take a lot of different forms. They can be an image or a 3D model or whatever it is. The problem is, where do you get that model. So if I am in my house and there is thousands of different objects, you're not going to have the 3D model most likely for the object that you want to pick up right now for the person. So there's this sort of data grab. But if you do have the model, it can be really, really accurate because you can know a lot about the object that you're trying to pick up. So the contribution of our approach is to try to bridge these two by enabling a robot to get the accuracy of the instance-based approach by autonomously collecting its own data that it needs in order to robustly manipulate objects. So we're going to get the accuracy of instance-based approach and the generality of category at the cost of not human time, but robot time to build this model. So here's what it looks like on our Baxter. It's going to make a point cloud. This is showing-- it's got a one pixel connect in its gripper. So you're seeing it doing a sort of raster scan. This is sped up to get a point cloud. Now it's taking images of the object. So it's got an RGB camera in its wrist. It's taking pictures of the object from lots of different perspectives. So the data looks like this. You segment out the object from the background. You get lots and lots and lots of images. You do completely standard computer vision stuff, SIFT and kNN, to make a detector out of this data. You can get a point cloud. This is the point cloud looks like at one-centimeter resolution. And after we do this, we're able to pick up lots of stuff. So this is showing our robot-- these two objects, localizing the object and picking things up. It's going to pick up the egg. And that's a practice EpiPen. There's a little shake to make sure it's got a good grasp. Now this works on a lot of objects, so let's see how it does on the ruler. So the way that the system is working is it's using the point cloud to infer where to grasp the object. But we don't really have a model of physics or friction or slippage. So it infers a grasp near the end because it fits in the gripper. It kind of looks like it's going to work. And it does fit in the gripper, but when we go and do that shake, what's going to happen is it's going to pop right out of the gripper because it's got this relatively low friction. There it goes and falls out. So that's bad, right? We don't really like it when our robots drop things. So before training, what happens is it falls out of the gripper. In the case of the ruler, there's sort of physics going on, right? Things are slipping out, and maybe we should be doing physical reasoning. I think we should be doing physical reasoning. I won't say "maybe" about that. But we're not doing it right now. And there's lots of reasons things could fail. So in other problematic objects, this is one of those salt shakers. It's got black handles that are great for our robot to pick up, but they're black, so they absorb the IR light, so we can't see them. So we can't figure out that we're supposed to grab there. That round bulb looks awesome. It's transparent though, so you get all these weird reflections. So the robots-- are inference algorithms is like, oh, that bulb, that's where we should pick it up. It doesn't fit in the gripper. So it will very often slip out of the gripper. So what our approach to solve this problem is, is we're going to let the robot practice. So we have-- I'm not going to go through the algorithm, but we have this unarmed bandit algorithm that lets us systematically decide where we should pick objects up. You can give it a prior on where you think good graphs are. And you can use whatever information you want in that prior. And if the prior was perfect, this would be boring. It would just work the first time, and life would go on. But if the prior's wrong for any reason, the robot will be able to detect it and fix things up and learn where the most reliable places are to pick up those objects. So here's an example of what happens when we use this algorithm. We practice picking up the ruler. I forget how many times it had. Maybe 20 on this particular object. One of the rifts on the algorithm is it decides when to stop. So we go a maximum 50 picks. But we might stop after three if all three of them work so that you can go on to the next object to train. So here it picks up in the middle and does a nice shake. OK, so what we're doing now is scaling up this whole thing. So this is showing our robot practicing on lots and lots of different objects. A lot of them are toys. My son likes to watch this video because he likes to see the robot playing with all of his toys. And I think playing is actually-- I mean it's one of those loaded cognitive science words, but I think that's an interesting way to think about what the robots are actually doing right now. It's doing little experiments trying to pick up these objects in different places and recording where it works and where it doesn't work. So this is sort of showing 16, 32, one in each hand objects being done in our initial evaluation. And at the end of this, basically, it works. So this is all the objects in our test set. And before learning, we were able to do with this proposal system, which uses the steps information, we get about 50% pick success rate. After learning, they go up to 75%. And the other really cool thing is that this is a bimodal distribution. So it doesn't say 75% is what you're going to get. A lot of these objects worked eight, nine out of 10 times or 10 out of 10 times. It goes from worst to best. So the good stuff is all over there, and the hard stuff is all over there. A lot of other objects were really hard. So that garlic press I think we picked it up one time. It's really, really heavy, so it slips out a lot. That gyro-ball thing has a lot of reflection, so we had trouble localizing it accurately. So we picked it up very few times. I think everything from about the EpiPen over was eight out of 10 or better. So not only-- so there's a lot of objects that we can pick up, and we can know which ones we can pick up. And which ones we can't. We are right now taking an aggressively instance-based approach. And the reason that we're doing that is I think there's something magic when the robot actually picks something up. So where I wanted to start is let's cheat in every way we can. Let's completely make a model that's totally specific to this particular object. But the next step that we're doing is to try to scale up this whole thing and then start to think about more general models to go back to that dream of category-based recognition. So if you look at computer vision success stories, one of the things that makes a lot of algorithms successful is data sets. And the size of those data sets is immense. A lot of the computer vision data sets, COCO DB from Microsoft, have millions of images, which are labeled with where the object is. But most of those images are taken by a human photographer on your cell phone or uploaded to Flicker. Wherever they got them from. And you get to see each object once. Maybe you see it twice from one perspective that a human carefully chose. You don't get to play with it. You don't get to manipulate it. In robotics, there's some data sets of object instances. The largest ones have a few hundred of objects. So computer vision people that I've talked to they laugh at it because it's just so much smaller compared to the data sets that we're working with. I think it's also so much smaller than what a human child gets to play with over the course of going from zero to two years old. I guess my son became a mobile manipulator around a year later around one and a half or so. I'm not sure exactly when. So one of my goals is to scale up this whole thing to change this data equation to be more in our favor. So there's about 300 of these-- this is the Baxter robot-- there's about 300 of them that Rethink Robotics-- Rod Brooks-- so we were talking about Rob Brooks in the previous talk. Rod founded this company Rethink Robotics. They've sold about 300 of them to the robotics research community. That's a very high penetration rate in robotics research. Everybody has a Baxter or a friend with a Baxter. So we're starting something which we're calling the million object challenge. And the goal is to enlist all of those Baxters, which are sitting around doing nothing a lot of the time-- to change this data equation. So what we're doing is we're going to try to get everybody to scan objects for us, so that we can get models, perceptual models, visual models, and also manipulation experiences with these objects to try to train new and better category models. And I think even existing algorithms may work way better simply because they have better data. But I think it also opens up the door to thinking about better models that we maybe couldn't even think about before because we just didn't have the data to play with them. So where we are right now is we've installed our stack at MIT on Daniela Rus's Baxter. That's this one. And we went down to Yale a couple of weeks ago to Scass's lab and we have our software on their Baxter. We're going to Rethink tomorrow. They're going to give us three Baxters that we're going to play with and install there. And I have a verbal yes from WPI. And a few other people have been like-- I pitched this at RSS. So a lot of people have said they were interested. I don't know if they'll actually translate to robot time. And our goal is to get about 500 or 1,000 objects between these three sites. Four sites I guess if the WPI gets on board. Four sites including us. So Rethink, Yale, MIT and us. And then do like a larger press release about the project. Advertise it, push over all of our friends with Baxters to help us scan. And then have yearly scanathons where you download the latest software and then spend a couple of days scanning objects for the glory of robotics or something. And really try to change this data equation for the better, so we can manipulate lots of things. So that's our plan for making robots that can robustly perform actions and real-world environments. More generally, I imagine like a mobile robot walking around your house at night and scanning stuff completely autonomously. Taking these pictures, building these models, hopefully, not breaking too much of your stuff. And not only learning about your particular house and the things that are in it, but also collecting data that will enable other robots to perform better over time. All right, so that's our attack on making robots robustly perform actions in real-world environments. So the next problem that I think is important for language understanding of human robot collaboration is making robots to carry out complex sequences of actions. So for example, this is this pantry again. There might be hundreds or thousands of objects that the robot could potentially manipulate. And it might need to do a sequence of 10 or 20 manipulations in order to solve a problem such as clean up the kitchen or put away the groceries. For work that I had done in the past on the forklift a lot of the commands that we studied and thought about were the level of abstraction of put the pallet on the truck. But one of our annotators-- we cleared the law of data on Amazon Mechanical Turk. And one of our annotators gave us this problem that I never forgot, which was how-- it was the actual forklift operator who worked in a warehouse, and he said if you paid me extra money, I'll tell you how to pick up a dime-- a dime, like a little coin with a forklift. Here's the instructions that he eventually without making us pay him gave us for how to solve this problem. So it was raise the forks 12 inches, line it in front of the dime, tilt it forward, drive a little bit over, you lower the fork on top of the dime, put it in reverse and travel backward, the dime kind of flips up backwards on top of the fork. Maybe you know how to drive a forklift, but you can see how that would work. And if you did know how to drive a forklift, you can follow those instructions and have it happen. But I knew that our system if we gave it these commands, there is no way that it would work. It would completely fall apart. And the reason that it would fall apart is that we gave the robot a model of actions at a different level of abstraction than this language is using. We gave it a very high-level of abstract actions, like picking stuff up and moving it into particular locations and moving things down. And if we gave it these low-level actions of like raising the forks 12 inches, the search steps that would be required to find a high-level thing like put the pallet on the truck would be prohibitively expensive. And I think if we want to have human-- but the thing is people don't like to stick at any fixed level of abstraction. People move up and down the tree freely. They give very high-level, mid-level, low-level commands. So I think we need new planning algorithms that support this kind of thing. So to think about this, we decided to look at a version of the problem in simulation. The simulator that we chose is a game called Minecraft. Five minutes, OK. So it's sort of this-- this is a picture from a Minecraft world. And we're trying to figure out new planning algorithms. So the problem here is that the agent needs to cross the trench. So he needs to make a bridge to get across the trench. So it's got some blocks that he can manipulate. And you have this combinatorial explosion of where the blocks can go. They can go anywhere. So in a naive algorithm, we'll spend a lot of time putting the blocks everywhere, which doesn't really make progress towards solving a problem. Whereas what you really need to do is focus on putting these blocks actually in the trench in order to solve the problem. Of course, on a different day, you might be asked to make a tower or make a castle or make a staircase, and then these might be good things to do. So you don't just throw out those actions. You want to have them both and figure out what to do based on your high-level goal. So we have some work about learning how to do this. So we have an agent that practices solving small Minecraft problems and then learns how to solve bigger problems from experience. This is showing transferring this from small problems to big problems in a decision theoretic framework, an MDP framework. And we've just released a couple of weeks ago a mod for Minecraft, the game called BurlapCraft. BURLAP is our reinforcement learning and planning framework that James MacGlashan and Michael Littman developed in Java. So you can run BURLAP inside the Minecraft JVM. Get the state of the real Minecraft world. Make small toy problems if you want. Or let your agent go in the real thing and explore the whole space of possible Minecraft spaces if you're interested in that simulation. OK, I'm almost out of time, so I'm not going to go too much into robots coordinating with people. But maybe I will show some of the videos about this work. The idea is that a lot of the previous work and language understanding shows people works in batch mode. So the robot does something-- the person says something, the robot thinks for a long time, and then the robot does something. Hopefully, the right thing. And as I said before, this is not how people will work. So we're working on new models that enable the robot-- this is a graphical model that shows how it-- talking about it in the car. What happens, it incrementally interprets language and gesture updating at very high frequencies. So this is showing the belief about which objects the person wants, updating from their language and gesture in an animated kind of way, right? Like, it's updating at 14 Hertz. So the idea is that the robot has the information. This is its own language. I would like a bowl. Both bowls go up. That one he points and then the one that he's pointing at goes up. So the robot knows very, very quickly, every time we get a new word from [INAUDIBLE] condition, every time we get a new observation from the gesture system, we update our belief. And just a couple of weeks ago, we had our first pilot results showing that we can use this information to enable the robot to produce real-time feedback that increases the human's accuracy at getting the robot to select the right object. This is some quantitative results. I'll skip it. OK, so that's the three main thrusts that I'm working on in my research group. Trying to make robots that can robustly perform actions in real-world environments. Thinking about planning in a really large state action spaces that result when you have a capable and powerful robot. And then thinking about how you can make the robot coordinate with people so that they can figure out what to do in these really large state actions spaces. Thank you.
MIT_RES9003_Brains_Minds_and_Machines_Summer_Course_Summer_2015	Lecture_82_John_Leonard_Mapping_Localization_and_Self_Driving_Vehicles.txt	The following content is provided under a Creative Commons license. Your support will help MIT OpenCourseWare continue to offer high quality educational resources for free. To make a donation or view additional materials from hundreds of MIT courses, visit MIT OpenCourseWare at ocw.mit.edu. JOHN LEONARD: OK, thanks. Thanks for the opportunity to talk. So hi, everyone. It's a great pleasure to talk here at MBL. I've been coming to the Woods Hole Oceanographic Institution for many years as my first thing over here at MBL. And so I'm going to try to cover three different topics, which is probably a little ambitious on time. But there's so much I'd love to say to you. I want to talk about self-driving cars. And use it as a context to think about questions of representation for localization and mapping, and maybe connect it into some of the brain questions that you folks are interested in, and time permitting, at the end mention a little bit of work we've done on object-based mapping in my lab. So my background-- I grew up in Philadelphia. Went to UPenn for engineering. But then went to Oxford to do my PhD at a very exciting time when their computer vision and robotics group was just being formed at Oxford under Michael Brady. And then I came back to MIT and started working with underwater vehicles. And that's when I got involved with Woods Hole Oceanographic Institution. And I was very fortunate to join the AI lab back around 2002, which became part of CSAIL. And really, I've been able to work with really amazing colleagues and amazing robots in a challenging set of environments. So autonomous underwater vehicles provide a very unique challenge because we have very poor communications to them. Typically, we use acoustic modems that might give you 96 bytes if you're lucky every 10 seconds to a few kilometers range. And so we also need to think about the sort of constraints of running in real time onboard a vehicle. And so the sort of work that my lab's done-- the more we investigate more fundamental questions about robot perception, navigation, and mapping, we also are involved in building systems. So this is a project I did for the Office of Naval Research some years ago using small vehicles that would reacquire mine-like targets on the bottom for the Navy. And so this is an example of a more applied system where we had a very small resource-constrained platform. And the sort of work we did is a robot built a map as it performed its mission, and then matched the map against the prior map to do terminal guidance to a target. Another big system I was involved with, as Russ mentioned, was the Urban Challenge. And I'll say a bit about that in the context of self-driving cars. So let's see. So who's heard any of the recent statements from Elon Musk from Tesla? So he said self-driving cars are solved- he said. And a particular thing that he said that just made my-- I don't know, maybe steam came out of my head-- was that he compared autonomous cars with elevators that used to require operators but are now self-service. So imagine you getting in a car, pressing a button, and arriving at MIT in Cambridge 80 miles away, navigating through the Boston downtown highways and intersections. And maybe that will happen. But I think it's going to take a lot longer than folks are saying. And some of that comes from just fundamental questions and intelligence and robotics. So in a nutshell, when Musk says that self-driving is solved I think he's wrong, as much as I admire what Tesla and SpaceX have done. And so to talk about that, I think we need to be very honest as a field about our failures as well as our successes, and try to balance what you hear in the media with the reality of where I think we are. And so I wanted to quote verbatim what Russ said about the robotics challenge, about a project that was so exhausting and just all-consuming and so stressful, yet so rewarding. So we did this in 2006 and 2007-- my wonderful colleagues, Seth Teller, John Howe, Amelia Fratoli-- amazing students and postdocs. We had a very large team. And we tried to push the limit on what was possible with perception and real-time motion planning. So our vehicle built a local map as it traveled from its perceptual data, using data from laser scanners and cameras. And we didn't want to blindly follow GPS. We wanted the car to make its own decisions because GPS navigation was part of the original quest with the challenge. And so Seth Teller and his student, Albert Wang, developed a vision-based perceptual system where the car tried to detect from curbs and lane markings in very challenging vision conditions. For example, looking into the sun, which you'll see in a second-- really challenging situation for trying to perceive the world. And so our vehicle-- at the time, we went a little crazy on the computation. We had 10 blades, each with four cores-- 40 cores-- which may not seem a lot now, but we needed 3.5 kilowatts just to power the computer at full tilt. We fully loaded the computer with a randomized motion planner, with all these perception algorithms. We had a Velodyne laser scanner on the roof. And about 12 other laser scanners, 5 cameras, 15 radars, and we really pushed the envelope on algorithms. And so when faced with a choice in a DARPA challenge, if you want to win at all costs you might simplify, or try to read the rules carefully, or guess the rule simplifications. But that would have meant just sort of turning off the work of our PhD students, and we didn't want to do that. So at the end of the day, all credit to the teams that did well. Carnegie Mellon-- first, $2 million, Stanford-- second, $1 million, Virginia Tech-- third, half a million dollars, MIT-- fourth, and nothing for fourth place. But it was quite an amazing experience. And in the spirit of advertising our failures I think I have time to show this. This used to be painful for me to watch. But now I've gotten over it. This is our-- [VIDEO PLAYBACK] - Let's check in once again with the boss. JOHN LEONARD: Even though we finished the race, we had a few incidents so DARPA stopped things and let us continue. - --across the line. JOHN LEONARD: Carnegie-Mellon, who won the race. Why did that stop? Let's see. - --at the end of mission two behind Virginia Tech. Virginia Tech got a little issue. [INAUDIBLE] Here's-- JOHN LEONARD: We were trying to pass Cornell for a few minutes. - Looks like they're stopped. And it looks like they're-- that the 79 is trying to pass and has passed the chase vehicle for Skynet, the 26 vehicle. Wow. And now he's done it. And Talos is going to pass. Very aggressive. And, whoa. Ohh. We had our first collision. Crash in turn one. Oh boy. That is, you know, that's a bold maneuver. [END PLAYBACK] JOHN LEONARD: So what actually happened? So it turned out Cornell were having problems with their actuators. They were sort of stopping and starting and stopping and starting. And we had some problems. It turned out we had about five bugs. They had about five bugs that interacted. And here's a computer's eye-- sort of, brain of the robot's view. Now back in '07, we weren't using a lot of vision for object detection and classification. So with the laser scanner-- the Cornell vehicle's there. It has a license plate. It has tail lights. It has a big number 26. It's on the middle of a road. We should know that's a car. Stay away from it. But to the laser scanner it's just a blob of laser scanner data. And even when we pull around the side of the car we weren't clever enough with our algorithms to fill in the fact that it's a car. And you have the problem when it starts moving of the aperture problem-- that as you're moving, and it's moving, it's very hard to tell and deduce the true motion. Now, another thing that happened was we had a threshold. And so in our 150,000 lines of code our wonderfully gifted student, who's now a tenured professor at Michigan, Ed Olson, had a threshold of 3 meters per second. So anything moving faster than 3 meters per second could be a car. Anything less than 3 meters per second couldn't be a car. Now that might seem kind of silly. But it turns out that slowly moving obstacles are much harder to detect and classify than fast moving obstacles. That's one reason that city driving or driving, say, in a shopping mall parking lot is actually in many ways more challenging than driving on the highway. And so despite our best efforts to stop at the last minute, we steered into the car and had this little minor fender bender. But one thing that we did is we made all our data available open source. And we actually wrote a journal article on this incident and a few others. And so if you'd asked me then in 2007, I would have said we're a long way from turning your car loose on the streets of Boston with absolutely no user input. And the real challenge is our uncertainty and robustness and developing robust systems that really work. But for our system, some of the algorithm progress we made-- I mentioned the lane tracking. Albert Wang, who's now, I think, working at Google, developed-- was given very sparse-- I'd say about 10% of the recent graduates or more are working at Google these days. AUDIENCE: Albert's at [INAUDIBLE].. JOHN LEONARD: Oh. OK. And then here is a video for the qualifying event to get into the final race. We had to navigate-- whoops, I can't press the mouse. That's going to stop. So we had to navigate along a curved road with very sparse waypoints. And so, in real time the computer has to make decisions about what it sees. Where is the road? Where am I? Are there obstacles? And there are no parked cars in this situation, but other stretches had parked cars. And our car-- in a nutshell, if our robot became confused about where the road was it would stop. It would have to wait and get its courage up, like lowering its thresholds as it was stuck. But we were the only team to our knowledge to qualify without actually adding waypoints. So it turns out the other top teams, they just went in with a Google satellite image and just added a breadcrumb trail for the robot to follow, simplifying the perception. So this was back in '07. Now let's fast forward to 2015. And right now-- so of course, we have the Google self-driving car which has just been an amazing project. And you've all probably seen these videos, each with millions of hits on YouTube. The earlier one of taking a blind person for a ride to Taco Bell, this was driving-- that was 2012, city streets in 2014, spring 2015. And then the new Google car, which won't have a steering wheel in its final instantiation, won't have pedals. It will just have a stop button. And that's your analogy to the elevator. And so I think that the Google car is an amazing research project that might one day transform mobility. But I do think, with all sincerity-- so I rode in the Google car last summer. I was blown away. I felt like I was on the beach at Kitty Hawk. It's like this just really profound technology that could in the long term have a very big impact. And I have amazing respect for that team-- Chris Urmson, Mike Montemerlo, et cetera. But I think in the media and in others, the technology has been a bit overhyped, and it's poorly misunderstood. And a lot of it goes down to how the car localizes itself, how it uses prior maps, and how they simplify the task of driving. And so even though people like Musk have said driving is a solved problem, I think we have to be aware that just because it works for Google, doesn't mean it'll work for everybody else. So critical differences between Google and, say, everyone else. And this is with all respect to all players. I'm not trying to criticize. It's more just trying to balance the debate. The Google car localizes on the left with a prior map, where they map the lighter intensity off of the ground surface. And they will annotate the map by hand-- adding pedestrian crossings, adding stoplights. They'll drive a car around many, many times, and then do a SLAM process to optimize the map. But if the world changes, they're going to have to adapt to that. Now, they've shown the ability to do response to construction, bicyclists with hand signals. When I was in the car we crossed the railroad tracks. That just blew me away. I mean, it's pretty impressive capability but more a vision-based approach that just follows the lane markings. If the lane markings are good, everything's fine. In fact, Tesla either just have released-- or are about to release-- their autopilot software, which is an advanced lane keeping system. And Elon Musk, a few weeks ago, posted on Twitter that there's one last corner case for us to fix. And apparently he-- on part of his commute in the Los Angeles area there is well defined lane markings. And part of it is a concrete road with weeds and skid marks and so forth. And he said publicly that the system works well if the lane markings are well-defined. But for more challenging vision conditions like looking into the sun it doesn't work as well. And so the critical difference is if you're going to use the LiDAR with prior maps, you can do very precise localization down to less than 10 centimeters accuracy. And the way I think about it is robot navigation is about three things-- where do you want the robot to be? Where does the robot think it is? And where really is the robot? And when the robot thinks it's somewhere, but it's really somewhere different, that's really bad. That happens. We've lost underwater vehicles and had very nervous searches to find them-- luckily-- when the robot made a mistake. And so with the Google approach they really nail this "where am I" problem-- the localization problem. But it means having an expensive LiDar. It means having accurate maps. It means maintaining them. One critical distinction is between level four and level three. These are definitions of autonomy from the US government-- from NTSA. A level four car is what Google are trying to do now, which is really, you just-- you could go to sleep. The car has a 100% control. You couldn't intervene if you wanted to. You just press a button. Go to sleep. Wake up at your destination. Musk has said that he thinks within five years you can go to sleep in your car, which to me I just-- five decades would impress me, to be honest. But level three is when the car is going to do most of the job, but you have to take over if something goes wrong. And for example Delphi drove 99% of the way across the US in spring of this year, which is pretty impressive. But 50 miles had to be driven by people-- getting on and off of highways and city streets. And so there's something about human nature, and the way humans interact with autonomous systems, that it's actually kind of hard for a person to pay attention. Imagine if 99% of the time the car does it perfectly. But 1% of the time it's about to make a mistake, and you have to be alert to take over. And research experience from aviation has shown that humans are actually bad at that. And another issue is-- and this is-- I mean, Mountainview is pretty complicated-- lots of cyclists, pedestrians, I mentioned the railroad crossings, construction. But in California they've had this historic drought. And most of the testing has been done with no rain, for example, and no snow. And if you think about Boston and Boston roads, there are some pretty challenging situations. And so for myself, when I first-- a couple of years ago I said I didn't expect a taxi in Manhattan in my lifetime-- a fully autonomous taxi-- to go anywhere in Manhattan. And I got criticized online for saying that. So I put a dash cam on my car, and actually had my son record cell phone footage. The upper left is making a left turn near my house in Newton, Mass. And if you look to the right, there's cars as far as the eye can see. And if you look to the left, there's cars coming at pretty high rate of speed, with a mailbox, and a tree. And this is a really challenging behavior for a human, because it requires making a decision in real time. We want very high reliability in terms of detecting the cars coming from the left. But the way that I pulled out is to wave at a person in another car. And those sort of nods and waves-- they're some of the most challenging forms of human-computer interaction. So imagine vision algorithms that could detect a person nodding at you from the other direction. Or here's another situation. This is going through Coolidge Corner in Brookline. And I'll show a longer version of this in a second. But the light's green. And see here-- this police officer? So despite the green light, the police officer just raises their hand, and that means the signal to stop. And so interacting with crossing guards and people-- very challenging, as well as changes to the road surface and, of course, adverse weather. And so here's a longer sequence for that police officer. First of all, you'll see flashing lights on the left-- which may be flashing lights, you should pull over. Here you should just drive past them. It's just the cop left his lights on when he parked his car. But the light's red. And this police officer is waving me through a red light, which I think is a really advanced behavior. So imagine a car that's-- imagine the logic for OK, stop at red lights unless there's a police officer waving you through it, and how you get that reliable. And now we're going to pull up to the next intersection, and this police officer is going to stop at a green light. And so despite all the recent progress in vision, things like image labeling, ImageNet-- most of those systems are trained with vast archives of images from the internet where there's no context. And they're so challenging for even humans to classify. So that if you had some data sets, like the Caltech pedestrian data set, if you got 78% performance, that's really good. But we need 99.9999% or better performance before we're going to turn cars loose in the wild in these challenging situations. Now going back more to localization and mapping. Here I collected data for about three or four weeks of my commuting. This is crossing the Mass. Ave. Bridge going from Boston into Cambridge. And the lighting is a little tricky. But tell me what's different between the top and the bottom video. And notice, by the way, how close we come to this truck. The slightest angular error in your position estimate, really bad things could happen. But the top-- this is a long weekend. This is Veterans Day weekend. They repaved the Mass. Ave. Bridge. So on the bottom, the lane lines are gone. And so if you had an appearance-based localization algorithm like Google's, you would need to remap the bridge before you drove on it. But the lines aren't there yet. And how well is it going to work? And so, this is just a really tricky situation. And, of course, there's weather. Now, snow is difficult for things like traction and control. But for perception, if you look at how the Google car actually works-- if you're going to localize yourself based on precisely knowing the car's position down to centimeters so that you can predict what you should see, then if you can't see the road surface you're not going to be able to localize. And so this is just a reminder of the sorts of maps that Google uses. So I think to make it to really challenging weather and very complex environments, we need a higher level understanding of the world. I think more a semantic or object-based understanding of the world. And then, of course, there's difficulties in perception. And so what do you see in this picture? The sun? There's a green light there. I realize the lighting is really harsh, and maybe you could do polarization or something better. But does anyone see the traffic cop standing there? You can just make out his legs. There's a policeman there who gave me this little wave, even though I was sort of blinded by the sun. And he walked out and put his back to me and was waving pedestrians across, even though the light was green. So a purely vision-based system is going to just need dramatic leaps in visual performance. So to wrap up the self-driving car part, I think the big questions going forward-- technical challenges, maintaining the maps, dealing with adverse weather, interacting with people-- both inside and outside of the car-- and then getting truly robust computer vision algorithms. We want to get in a totally different place on the ROC curves, or the precision recall curves, where approaching perfect detection with no false alarms. And that's a really hard thing to do. So I've worked my whole life on the robot mapping and localization problem. And for this audience I wanted to just ask you a little question. Does anyone know what the 2014 Nobel Prize in medicine or physiology was for? Anybody? AUDIENCE: [INAUDIBLE] AUDIENCE: Grid cells. JOHN LEONARD: Grid cells. Grid cells and place cells. And so this has been called SLAM in the brain. Now, you might argue. And we might be very far from knowing. But I think it's just really exciting to-- so for myself, I'll explain. I've had what's called an ONR MURI grant-- multidisciplinary university research initiative grant-- with Mike Hasselmo and his colleagues at Boston University. And these are a couple of Mike's videos. And so, I think Matt Wilson spoke to your group. And the notion that in the entorhinal cortex that there is this sort of position information that's very metrical, and it seems to be at the heart of memory formation, to me is very powerful and very important. And so, we have this underlying question of representation. How do we represent the world? And I believe location is just absolutely vital to building memories and to developing advanced reasoning in the world. And the fact that grid cells exist-- to me-- and they have this role in memory formation is just this really exciting concept. And so, in robotics we call the problem of how a robot builds a map and uses that map to navigate, SLAM-- simultaneous localization and mapping. This is for a PR2 robot being driven around the second floor of our building, not far from Patrick's office if you recognize any of that. And this is using stereo vision. My PhD student, Hordur Johannsson, who graduated a couple of years ago, created a system to do real time SLAM and try to address how to get temporally scalable representations. And one thing you'll see as the robot goes around occasionally is loop closing, where the robot might come back and have like, an error and then correct that error. So this is the part of the SLAM problem that in some ways is well understood in robotics, which is how you detect features from images, track them over time, and try to bootstrap up, building a representation and using that to locate your estimation. And I've worked on this my whole career. And as a grad student at Oxford, I had very primitive sensors. So for a historical SLAM talk I recently digitized an old video and some old pictures. This was in the basement of the engineering building at Oxford. This is just the localization part of how you have a map, and you generate predictions-- in this case for sonar measurements. And at the time there we had-- I'm sitting at a SUN workstation. To my left is something called a data cube, which for about $100,000 could just barely do like real time frame grabbing and then edge detection out. And so vision just wasn't ready. And the exciting thing now in our field is vision is ready-- that we're really using vision in a substantial way. But I think a lot about prediction. If you know your position, you can predict what you should see and create a feedback loop. And that's sort of what we're trying to do. And so SLAM is a wonderful problem, I believe, for addressing a whole great set of questions, because there are these different axes of difficulty that interact with one another. And one is representation. How do we represent the world? And I think that question-- we still have a ton of things to think about. Another is inference. We want to do real time inference about what's where in the world and how we combine it all together. And finally, there's a systems in autonomy access, where we want to build systems, and deploy them, and have them operate robustly and reliably in the world. So in SLAM, here's an example of how we pose this as an inference problem. This is from the classic Victoria Park data set from Sydney University. A robot drives around, in this case, a park with some trees. There are landmarks shown in green. The robot's positioner drifts over time. We have dead reckoning error. That's shown in blue. And we estimate the trajectory of the robot in red, and the position of the landmarks from relative measurement. So as you take relative measurements, and you move through the world, how do you put that all together? And so we, cast this as an inference problem where we have the robot poses, the odometric inputs, landmarks-- you can do it with or without landmarks-- and measurements. And an interesting thing-- so we have this inference problem on a belief network. The key thing about SLAM is it's building up over time. So you start with nothing and the problem's growing ever larger. And, let's see, if I had to say-- 25 years of thinking about this up through 2012, the most important thing I learned is that maintaining sparsity in the underlying representation is critical. And, in fact, for biological systems I wonder if there is evidence of sparsity. Because sparsity is the key to doing efficient inference when you pose this problem. And so many algorithms have basically boiled down to maintaining sparsity and the underlying representations. So just briefly, the most important thing I learned since then in the last few years-- I'm really excited by building dense representations. So this is work in collaboration with some folks in Ireland-- Tom Whelan, John McDonald-- building on KinectFusion from Richard Newcombe and Andrew Davison-- how you can use a GPU to build a volumetric representation, and build rich, dense models, and estimate your motion as you go through the world. So this is something we call continuous or spatially extended KinectFusion. This little video here from three years ago is going on in an apartment in Ireland. And I'll show you the end result. Just hand-carrying a sensor through the world-- and you can see the quality of the reconstructions you can build, say, in the bathroom, the sink, the tub, the stairs, to have really rich 3D models that we can build and then enable the more advanced interactions that Russ showed. That's fantastic. And I mentioned loop closing-- something we did a couple of years ago was adding loop closing to these dense representations. So this is-- again, in CSAIL-- this is walking around the Stata Center with about eight minutes of data going up and down stairs. If you watch the two blue chairs near Randy Davis's office, you can see how they get locked into place as you correct the error. So this is taking mesh deformation techniques from graphics and combining it. So the underlying pose graph representation is like a foundation or a skeleton on which you build the rich representation. OK. So this is the end resulting map. And there's been some really exciting work just this year from Whelan and from Newcombe in this space of doing deformable objects, and then really scalable algorithms where you can sort of paint the world. So the final thing I want to talk about in my last few minutes is our latest work of using object-based representations. And for this audience, I think if you go back to David Marr, who I feel is unappreciated in the historical sense of how I feel, that vision is the process of discovering from images what is present in the world and where it is. And to me, the what and where are coupled. And maybe that's been lost a bit. And I think that's one way in which robotics can help, I think, with vision and brain sciences. I think we need to develop object-based understanding of the world. So instead of just having representations that are a massive amount of points or purely appearance, where we can start to build higher level and symbolic understanding of the world. And so I want to build rich representations that leverage knowledge of your location to better understand where objects are and knowledge about objects to better understand your location. And just as a step in that direction, my student, Sudeep Pallai, who was one of Seth's students, has an RSS paper where we looked at coupling using SLAM to get better object recognition by effectively-- so here's an example of an input data stream from Peter Fox's group. There's just some objects on the table. I realize it's a relatively uncluttered scene. But this has been a benchmark for RGBD perception. And so, if you combine data as you move from the world using a SLAM system to do 3D reconstruction on the scene, and then using the reconstructed points to help improve the prediction process for object recognition, it leads to a more scalable system for recognizing objects. And it comes back to this notion to me that a big part of perception is prediction-- the ability to predict what you see from a given location. And so what we're doing is we're leveraging off techniques and object detection, featuring coding and the newer SLAM algorithms, and particularly the semi-dense orb SLAM technique from Zaragoza, Spain. And so I'm just going to jump to the end here. The key concept is that by combining SLAM with object detection we get much better performance and object recognition. So on the left shows our system. On the right is a classical approach just looking at individual frames. And you can see, for example, here, the red cup that's been misclassified would get substantially better performance by using location to cue the object detection techniques. All right. So I'm going to wrap up. And just a little bit of biological inspiration from our BU collaborators, Eichenbaum has looked at the what and the where pathways in the entorhinal cortex. And there's this duality between location-based and object-based representations in the brain. And I think that's very important. OK. So my dream is persistent autonomy and lifelong map learning and making things robust. And just for this group I made a-- I just want to pose some questions on the biological side, and I'll stop here. So some questions-- do biological representations support multiple location hypotheses? Even though we think we know where we are, robots are faced with multimodal situations all the time. And I wonder if there is any evidence for multiple hypotheses in the underlying representations in the brain, even if they don't rise to the conscious level, and how experiences build over time. And the question-- what are the grid cells really doing? Are they a form of path integration? Or there obviously, to me, seems to be some correction. And my crazy hypothesis as a non-brain brain scientist is, do grid cells serve as an indexing mechanism that effectively facilitates search-- so a location index search so that you can have these pointers to what and where information get coupled together.
MIT_RES9003_Brains_Minds_and_Machines_Summer_Course_Summer_2015	Tutorial_51_Tomer_Ullman_Church_Programming_Language_Part_1.txt	The following content is provided under a Creative Commons license. Your support will help MIT OpenCourseWare continue to offer high-quality, educational resources for free. To make a donation or view additional materials from hundreds of MIT courses, visit MIT OpenCourseWare at ocw.mit.edu. TOMER ULLMAN: And today, with your active help and participation, I hope to run a probabilistic programming tutorial in the time that we have left. And we're going to focus specifically on a language called Church, which is a probabilistic programming language that was developed in Josh Tenenbaum's Group, but it's now taken on a life of its own and has set up shop in other places. Before I get started, I should say, I was, sort of, looking for a good image. I didn't like a blank page, so I was googling just Church tutorial. This is the first thing I found. It's an image for Minecraft about how to build a church in Minecraft. Does any of us-- people have heard of Minecraft? AUDIENCE: Yeah. TOMER ULLMAN: They've played with Minecraft? OK-- just in case you don't know, Minecraft is a sort of procedurally-generated world where you get some building blocks, literally, building blocks, that you can build stuff with. And you can build an infinite number of things, including a computer, and a church. And it's very cool. And I thought it's actually not that bad of an image for a tutorial about probabilistic programming language, which is also about, sort of, procedurally-generative things that use small building blocks to build up an entire world. And I thought, OK, that's the first hit I got. What's the other hit? Well, it's just another church, and another church, and another church, another church, another church that you can build in Minecraft from many different angles and many different tutorials, so maybe, instead of this, you can just train some deep-learning algorithm to, I don't know, learn a billion churches and do that. That's not what we're after. So probabilistic programming, Josh already talked a bunch about this, so I'll sort of be repeating him, or channeling him. It's about combining the best of both worlds in the, sort of, two states of AI right now, which is statistical modeling and logic. And in many models, you have this, sort of, dual question of representation and learning. And it's really, sort of, a problem for cognitive science, going back to the days of before cognitive science, right? I mean, this is the sort of problem that a lot of people had when they tried to model the human mind. This goes back to Turing. Sort of, when we want to build a system that is human-like in its intelligence, the two questions that we face are, what are the representations that it will have, and how is it going to learn them or how is it going to learn anything new? And you often have to, sort of-- it's a short blanket problem, right? If you try to cover your head, your feet are sort of not getting anything. If you try to cover your feet, your head's not getting anything. Because oftentimes, you find that, if you stick to a particularly easy representation that's sort of easy to code or rather something, a kind of a presentation that's easy to learn, like say a vector of weights that you're just trying to shift your weights around, then, yes, that might be easy, relatively easy, but you're sort of stuck with the representation that you can learn are weights. Or Josh was making a big point about this, and it is a big point, that if you try to learn something like causal Bayes nets, then you're sort of limited by that representation. That is your representation of these sort of circles and arrows that go into other circles. And that might get you very, very far. And you might even have very good learning algorithms for those particular models, for those particular representations, that are tailored for those representations. Like, in these causal circles and arrows, belief propagation might be a very good learning algorithm, but if you commit to that representation, then you are sort of stuck with that representation. And you might not be flexible enough to learn all the stuff that you want to. And a very flexible representation, sort of, one of the more flexible ones that have come onto the scene in the past years is why don't we try to learn a program? I say it's come onto the scene in recent years. That's not exactly true. People have been interested in learning programs for many, many years, for many, many decades, but they sort of try to infer them from kind of a logical perspective, not really getting these probabilistic learning algorithms. I'm sort of throwing words out there, but it'll make more sense as I go through it. And you already have some of Josh's stuff to carry you through. But the point is, there's always these questions of learning and representation. For probabilistic programming languages, the representation is not circles and arrows, it's not vectors of weights, it is programs. That's what you're trying to learn. That's what you're trying to figure out the world with. And then there's a question of how do you learn these programs, but we'll get to that. OK, let's see, so we think it's a good representation for AI and cognition for all the reasons that Josh just talked about. And there's been a growing interest in these things for the past 10 years, witnessed both by the proliferation of many, many different types of programming languages-- sorry, probabilistic programming languages. I don't know whether to call them PPL, or people, or what exactly. But probabilistic programming languages, there's been PyMC based on Python, there's Church, which you're going to play with right now, but also BLOG, WinBUGS, ProbLog, Venture, many others that I haven't mentioned here. So first of all, there's many of them. And also, DARPA has started taking interest and has given a large grant to advance this field. They think it might be big. If you're in, sort of, probabilistic programming more generally than Church, you think it's interesting to follow, you want to learn more about it, you should go to this thing, probabilistic-programming.org wiki. It sort of keeps it up-to-date with many, many, many, different types of programming languages. You don't necessarily have to write this down right now. I will send you the slides later, but just, sort of, keep it in mind form, link to it in your head. There's also this, sort of, nice summary from this DARPA-- so DARPA started sending this about a year ago. And someone already went to a summer school on probabilistic programming and, sort of, wrote the state of the field. It's six months ago. It's a bit outdated, but it also makes for an interesting read for those of you who want to follow that. OK, so that's about probabilistic programming, very, very, very generally. What about Church very, very, very generally? So as I said, Church is one example of a probabilistic programming language. It was developed by several people at MIT who have since gone on to do other different things like continue to develop Church at Stanford. That's Professor Noah Goodman. Although, of course, he's doing many, many other things. There's also been Vikash Mansinghka, who has gone on to develop other probabilistic programming languages like Venture at MIT. And one thing to say generally about probabilistic programming languages is that, usually they are based on an already existing language. So you take MATLAB and you try to make it probabilistic. You take Python and you try to make it probabilistic. Julia has a probabilistic programming implementation. Church in particular is based on Scheme, which is the derivative of LISP, which is itself sort of an attempt to capture lambda calculus, which is not a programming language, it is an approach to trying to think about all possible functions developed by Alonzo Church. And that's why Church is called Church. It has nothing to do with the actual buildings. So the point about Scheme which is very nice is that it's very compositional. And anything that you write can then be passed off into the other functions as the data. You'll see some examples of that. Church has several inference engines that you can try to run. We'll get into that. The backbone of it is Metropolis-Hastings-type sampling over possible programs, but it has other types of programming, including explicit enumeration. If your space is small enough, it can just look at all the possible ways to run a program. It has rejection query. Again, we'll get to this. Don't worry about, like, what is he talking about. Yeah, so it has a whole bunch of-- you know, particle filtering is one attempt at that. But the point is there are-- each probabilistic programming language has its own set of inference engine. Some of them try to go the Metropolis-Hastings route. Some of them try to say, well, it's a probabilistic programming language, but it's really limited to causal Bayes nets, so the inference engines are going to be stuff that's good for causal Bayes nets. But all of them sort of share this dream of, it's easier to write the forward model than the inference. And it's really annoying. Those of you who have ever tried to write an inference engine or to write inference over any sort of model, it's really annoying to write that. And it usually sort of only works for the one thing that you've built. And one of the selling points of probabilistic programming languages, one of the reasons that DARPA took an interest, beyond the fact that they can try to capture the human mind, and flexible AI, and all that, is they have this sort of promise, this pitch that, why don't you just write down the forward model, how you think the world works, and we'll, kind of, take care of inference for you. And in many cases, it turns out to be a lot easier to write the forward model than to try to write the inference engine for it. In fact, you can very quickly get to something that's even, like, five or six lines of code long, that would be intractable, would be very hard to write down the analytic expression for, would be very hard to think about what would be the inference engine for, but it's really just easy to write. I mean, all you have is a set of assumptions. And you're trying to figure out how they work together. Again, we'll see some examples of that. But my point was all probabilistic programming languages are about writing the forward model and then, sort of, trying to do the inference for you. Another point about Church in particular, it is under construction, so you'll notice this when you write it now. It will break. It will freeze. It will do all sorts of annoying things, so it is under construction. It's not exactly something that you would then go and work with like MATLAB. Let me put some caveats on that caveat, which is these two asterisks right here. First of all, despite being a, sort of, a toy language, it's already been used in several serious scientific papers, including a paper in Science, because it is very easy to make certain points about cognition or about computational cognition in Church that is very hard to do in certain other languages. In particular, things that require recursion, or inference over inference, where you write down sort of the way that you think about an agent, then you put that into another agent, that can be very hard to write in certain languages. Church can kind of do that more easily. Let's see, I had another caveat, which is-- what was that? Oh, another caveat is that, despite it being under construction, you sort of think, well, why should I worry about this thing? Why should I even bother hacking with it? Is because, you'll notice there's probmods.org. And there are just a ton, a ton of examples. There's a semester worth of examples of all sorts of things from both cognition, and AI, and interesting statistical models that are very easy to understand in Church. And for me at least, it was very much a process of demystification that something like this can help with. You learn about something like the Chinese restaurant process, the Dirichlet process, nonparametrics, and it's kind of hard to read the textbook description of it. It's hard to wrap your head around. And then you go and you write three lines of code, or five lines of code. And you think, oh, that wasn't so bad, right? And it's sort of easy to write a bunch of these things in Church, so it's a useful tool for demystification. It's a useful tool to get a handle on certain models in cognition and statistics, so those are the two asterisks. Be warned, but also, you know, do play around with it. Let's see, the founding paper, for those of you who are interested, you can look at this link later on. It was by Goodman, Mansinghka, Dan Roy, Bonawitz, and Tenenbaum. And for those of you who, by the way, have already read about Church a bit, you think that this tutorial is a bit-- maybe it was-- I should say, we'll start off very, very easy, OK? We'll do things like addition. We'll do things like flipping coins, OK? If you think that this is-- maybe you've already read through probmods, you've already done a few chapters of that, by all means, use this time to continue to think about probabilistic programming, for example, either by talking to me, and I'll find something for you, or by going to forestdb.org. Again, I'll give you that link for those of you who want it. It has a whole repository of different probabilistic programming models that you can play with, think about, see how you would change them, and things like that. Also after this tutorial, if you're still interested, you can go to that link. Oh and one last thing. There's sort of a-- you can't see that right there. One last thing that I should say about Church, it's based on Scheme. But a lot of the people that have sort of been doing a lot of work on it have become more in love with JavaScript. In fact, the thing that you're going to be working on is sort of a JavaScript implementation of Church under the hood. And they've started to implement something called WebPPL, so Web Probabilistic Programming Language. It's a language that's specifically a derivative of JavaScript. For those of you who like JavaScript, you can play with that. And if you go to WebPPL.org, if you search for WebPPL, again, I can leave you the link for that. It's sort of here, but you can't see it. There are, again, a lot of nice examples there of different programming language-- programs that you can write in JavaScript. OK, that was a very long-winded introduction, caveats, and setting up different things. The objectives for this tutorial is, first of all, to become familiar with the Church syntax, it can be a little wonky, if you don't know it, at first, to run forward a few models to give you an example of just, before inference, an example of, here's my forward model, here's how I describe the world, now let's try sampling from it. Let's sample, sample again, sample again, sample again, see what distributions we get. Get a sense for the point that I'm going to make a few times, which is once you write your forward model, that is a representation of a distribution-- and I'll come back to this point, but just, sort of, keep that in mind. You write down a program. And you run it forward. And you get a sample. You run it again and you get a different sample. You run it in the limit, you get some distribution. Some other constructs like memoization-- after we do all of this, we'll try to get at sampling, and the query operator, and really, conditioning and inference. So we said we'll try to run a few models forward. Once we do that, we'll try to get the hang of inference. So you'll try to write down a forward model about things like a coin, or goal inference, or things like that. And you'll try to actually infer something, like what is the weight of the coin, from some data, like some coin flips, some very simple stuff. OK, and we'll go through some examples, like, as I said, coin flipping, maybe causal networks, maybe intuitive physics and intuitive psychology. I do hope to get to intuitive psychology. We'll see if we get to that. So some prerequisites and set up, that's what I asked you to do at the beginning. If you happen to have a local implementation, you can open that now. If you didn't, just go to probmods.org/play-space.html and open that up. And we're going to play a game of Noisy Tomer Says. So now you should also-- open this, open a browser, go to that, or open your local implementation. Also open up the file that I sent you of-- it should have have been called, like, student copy, something like that. It contains a bunch of things that we're basically going to just sort of copy, paste into the browser. Now, the nice thing about this browser is, it is sort of a working implementation of Church. You just paste in the code. You hit run. It runs, OK? So you guys should all more or less have a screen like this. I'll take this out so I don't sit on it right now. Does everyone have more or less something like this, some sort of browser that you can type things into and press run? Over there? OK, we'll start off with some very, very simple stuff that you should already have in the syntax of the Church tutorial, so just try either pasting in or typing in things like this thing. So the first thing you'll notice is that, over here, it's what's called-- sorry, let me adjust this screen so it's not actually-- so that you can see it. Zone C over here, you should be looking-- I've sort of done over here, plus 2 2, and the result is 4. So the first thing to see, some of you may be familiar with this, who's somebody with Polish notation, where you just go plus 2 2? Instead of going 2 plus-- who is not familiar with Polish notation? OK, good, thank you. Polish notation just means that, instead of writing 2 plus 2, you write plus 2 2, so you write that the thing that operates, the function, outside, and you write all the arguments for the function like that. In fact, most of the time, you do this. When you write down functions for code, you usually write the function then the things that it operates on. But here, it's going to work for everything. And it can be a bit confusing at first when you do things like plus 2 2. The second thing is that you put brackets on anything that you want to evaluate, OK? So, for example, here is an expression. The expression is plus 2 2. And you want to evaluate that expression. So for example, I wanted to evaluate the expression-- I think I put some, like, cursor for-- so you can see what I'm doing with my thing. OK, if you want to do something like, you know, times 2 2, that would be the same thing. And I would go to run. And that would be, of course, 4 again. It let's you do some other examples from here. Like there's a bunch of simple logic, like you might do display. Display is just a way to run it, to-- sorry, to display the result over here. You can do a bunch of logic things, like equal. So again, the operator is outside. And you would do equal question mark 2 2, and then evaluate that expression. And you can do bigger than equals, all these different things. AUDIENCE: the question mark? TOMER ULLMAN: The question mark is just-- I've just named it that way. It doesn't actually have any sense. I could have just called it equal-- sorry, no, sorry. There is no particular meaning to the question mark. It's just that this thing, this operator, is called equal question mark. That's the name for it. And it's just-- it is the equals operator. That's how you check if two things are equal to one another. In languages like Python, you would do, you know, equals equals, like that. This is how you do it here, OK? Let's see, a few other simple syntax things. So you might say, for example, the statement for defining variables is, shockingly enough, define, so you would do define x 3. And now, the next time that I do x, then hopefully-- and I run that-- then it'll show 3. There are a few other basic syntax things, like lists, that might be important, like, you know, define x to be a list of 1 2 3. And if you run that, then you'll get 1 2 3. Again, we're starting out very, very slow, but we'll hopefully get soon to more things like Gaussian processes. Some simple things like if-then statements-- OK, I'm just copying and pasting off of this document that you should all have, so that's why I'm, sort of, running through it. But the point is that you would do-- the syntax for doing an if-then conditional statement is like this. You write down if, and then you write down the condition that either evaluates to true or to false. So it's if this condition, do the first thing. If it's false, do the second thing. In this particular case, I have defined a variable called socrates. I've defined it as drunk. And then I run the condition equal socrates drunk, if that's true, then return the answer true. Or, you know, I could have written return the answer, Socrates is a drunk. If it's false, return the answer false. Did everyone more or less get the conditional? It just says, if condition, return the first thing otherwise, the thing on the second line. Another important thing before we start getting at more things like recursion and forward sampling is the notion of, how would I define a function? So, so far we've defined variables, right? I could have defined something like define x 2, right? And then that would have just been that. But I want to define, probably, functions, so I might define something like define-- and now I have two options. There are two ways of defining functions in Church. One of them is to do the following. You define square. And then you say, well, square is, itself, a procedure. It is a lambda. And I'll explain this as I go along, just watch me, sort of, type it. It takes in a particular argument, say, x. And then what it does to, is it multiplies x by x. So the point is, you say, well, here, x is a particular thing. It is an object. What is it? It is just 2. Here, square is a thing. What sort of thing is it? It is this thing. Ah, what is this thing? This thing is a procedure that-- this is the only thing that you need to know about functions. Lambda is the thing that actually defines functions, OK? It is a procedure that takes in some number of arguments, in this case, just one argument. You could have called it anything. I just called it x. You could have called it argument1. You could have called it socrates. You could have called it fubar. But the point is, it takes in this argument. And then what does it do with it is the next thing? So you say, lambda, number of arguments that you take in. And then what do you do with it? In this case, you just do times x x. So this is a function called square, very basic stuff. It takes in an argument and it multiplies it by itself, so it is the square of x, x times x, very simple. There's another way of doing that if you don't want to type out lambdas, if you don't want to start doing lambda this, lambda that, it's sort of annoying. Let me just give you one more example. Like, if I wanted something with two arguments, I could have done-- you know, I could have called it something like my-proc lambda x y. And now, what it does is, it multiplies xy. OK, this is an example of a thing. What sort of thing is it? It is a procedure. I know it's a procedure because it starts with lambda. It takes in two arguments. Here they're called x and y. What does that do with it? It multiplies x times y. Really, this is just multiplication. So after I define this procedure, I could then do, like, my-proc-- sorry, I should have explained that. Then you do my-proc, say, 2 8, or something like that. AUDIENCE: [INAUDIBLE] TOMER ULLMAN: Yeah, sorry-- that's a very good question. And it would bring back 16. Sorry, once I define my thing, this is an operator now. This is an operator that can be applied to arguments. And you apply it by doing that parentheses that we just saw. If I just tried, by the way, like, without applying it, if I just tried something like this, what you would get back is, it would say, this is a function, because it just says, what is this thing? You try to evaluate it. You're not evaluating on anything, so it just returns, what is this thing? It's a function. It's a function that expects x y and then multiplies them. If you actually want to apply it on something, you would need to provide with some input arguments. So I said, let's try to define square as a lambda of x. That does-- it takes in an x and multiplies x by x. There's one more way to define a function, which, it sort of gets rid of this lambda type thing. It's exactly equivalent to the thing I just showed you, it just takes a bit less writing, which is to say, define-- I just misspelled square, didn't I? Yes. Define square x-- like that-- times x x. Now what this is saying, so this just goes straight to saying, like, before I would say, define this thing 2. OK, and then I said, define this thing, the square, as this procedure. Here you can say, I want to directly define a procedure. I'm not going to bother with this lambda stuff. I want to directly define a function. I want to directly define a procedure. Can I do that? Yes, you could if you wanted to. You would just directly put these brackets right there. You would say define. And if the next thing is some brackets, then it says, OK, I'm going to define a procedure where the name of the procedure is square. And it takes in one argument, which is x. And what it does to it is times x x. And if you do it that way, then under the hood, what Scheme does is actually writes it out like this. It puts in the lambda where it expects, but again, this is not terribly important stuff. And those of you are, sort of, tuning out, and saying, well, fine. And you just wanted to learn about-- a bit more about how probabilistic programming works, don't worry. We'll get to some examples in about 10 minutes. Here's another very useful thing that you might want to do in many of your things. This is called the map. And the way map works is, you map a function to a bunch of arguments. So you would say-- map is just a high-level function which takes in a particular procedure. Then it applies it to each one of these things individually, OK? So square, in this case, as we said, it is a thing that takes in one argument. So this is now going to take square and apply it to 1. So then I'm going to take square and apply it to 2, take square and apply it to 3. And the result of this is just going to be a list of squares, 1 4, 9, 16, 25, simple enough? Yes. But map is very useful. You should probably know about it. OK, some simple things like, recursion, OK, so suppose I wanted to apply square to the list from 1 to 100, and suppose I didn't have the range 1 to 100. Most languages in Scheme actually does have something called range, which gives you all the numbers from 1 to 100. Suppose I didn't. Suppose I want to construct all the numbers 1 to 100. I don't want to actually write them down-- 1, 2, 3, 4, 5, 6, all the way up to 100. I can write down something that does that. And it uses a little bit of recursion. And the way it does it is this. This is just to get you used to recursion, because we'll be seeing it a little bit later. And this says, OK, I'm going to define something called range, which takes in an argument-- you should now be used to it, this is the same thing that we defined over here. We're going to call something a procedure. And we're going to call-- we're going to define a procedure. It's called range. It takes in an argument, n, one argument. What does it do? Well, it depends. It does a conditional. A conditional, it depends, let's see, is n equal to 0? If it's 0, just give me back an empty list. Does everyone sort of see that, if equal n 0, give me back a list. What if it's not 0? What if I did range 10? Oh, well, in that case, append-- another thing that you might want to know, so it's just combine these two things-- append what with what? Append range again with n minus 1 and with n. The point here is to say, OK, how do I get the numbers 1 to 100? I just, sort of, say range-- I want the range 1 to 100, so I say, 100-- am I at 0 yet? No, so take 100 and append it with range 99. What does range 99 do? Well, is 99 0? No, so give me back 99 plus what? Plus range 98. Is 98 0? No, keep going, so it's basically recursing-- range is a recursive function that calls itself until it hits 0, very simple recursion. And now you can do this to write out all the numbers from 1 to 100. And then you, if you were so inclined, you could do math square to that. OK, and we run that. And it gives me all the numbers from-- the squares of the numbers from 1 to 100. So far we've talked just about very basic stuff. This is no different from Scheme. You are all experts in Scheme notation and things like that. Let's move on to something a little bit more interesting that Church can do, which is, for example, take random sequences, and it can take random-- how should I put this? Kind of like plus is a basic thing in certain programming languages, it's a primitive, right? It's written into the language what plus means, what times means. You don't have to define that. The way most languages work is that they have this sort of long list of things that they need to evaluate. And they start evaluating them. And they're, sort of, OK, did I hit an expression I know, like a number or not? And it's, sort of, no, you didn't hit it yet. OK, fine, keep evaluating, keep evaluating, keep evaluating until you get some sort of primitive. And a primitive procedure could be something like plus or a number. In Church, there are primitive procedures which are random primitive procedures. They are procedures that, when you hit them, what you do is, you just return a value, a sampled value, from this expression, from this probability distribution. So the most basic random primitive, the most basic distribution that you can do in Church is something called flip. And if you just write down flip in Church, what you'll get, if you run it like that, is it tells you, well, it's a function. And it depends on certain arguments. And it tells you many, many things about it, but that's not what we want. We want to evaluate it, so put some parentheses around it. And we'll run it. And it will give us back false. OK, let's try that again. So let's run that again. It will gave us back true, OK, interesting. And if we run that again, you know, we get false. We run it again, and we get maybe true, maybe false. You could do repeat 1,000 times flip. OK, repeat is another important thing that you would need to know. It just says repeat as many times as you want to repeat some sort of function. In this case, the function is flip. OK, so repeat flip 1,000 times. I hope you guys are trying this while I'm saying this. Are people trying this more or less? OK, cool. So repeat 1,000 times flip. And what you'll get back is this long list of true, false, true, false, false, true, false, true. And it's independent from one another, because it's an exchangeable random sequence. And if you want to see what this looks like, well, you could just do something like hist. And you would run that. And you would get, you know, more or less 50-50. Not exactly 50-50, because I only ran it 1,000 times. If I had run this in the limit, what I would get is 50-50 on true-false. Now, what's nice about this is that this sort of gets at this thing that I was talking about earlier, where there's dual representation for any sort of probability distribution. You could either write the probability distribution in math. You could sort of say, well, the probability of true is 0.5. And the probability of false is 0.5. Now I've defined a distribution in math. And now you can say, well, what's conditioned on this, what can you do, and things like that. Or what you can do is, you can write a program such that, when you run it, it will sample one of these values. And in the limit, it samples it's such that it approximated the thing that we just defined in math. And you might say, well, why not just define it in math? Because oftentimes, it gets very, very, hairy very, very fast. And in fact, any sort of probability distribution that's well-defined and well-behaved, you can write as a program. A program which, if you run it many times, its sampling profile, the thing it will give you back if you sample it many, many different times, will give you back that probability distribution. Or you could equivalently say that, what it means for a probability distribution to be a probability distribution is to be some sort of program, to be some sort of procedure that gives you back a sample. And in the limit, you get some sort of thing that we're going to call the probability distribution. Actually, that's the way we define the probability distribution. And again, this gets in-- so one way to think about Church programs is that any Church program that you write-- if you just write plus 2 2, you'll get back 4. That's, in a way, a deterministic program, right? The probability of getting back 4 on this execution equals 1, but there are many other things that you could write and you could get back interesting things for them. And the point is to write something like a generative model that describes some sort of thing about the world. And when you run it forward, you get to a certain sample, but if you run in many, many different times, it gives you the probability distribution that this model describes. And now, if you-- and again, I'm getting slightly ahead of myself. If you change that model, if you, for example, condition on something, you'll get a different model. You'll get a different program. And you're trying to find the program such that its output will match the data. OK, but let's back up a little bit. And we're still in flip land. So we have here something which is flip. That's very, very basic. Flip can also be-- AUDIENCE: [INAUDIBLE] TOMER ULLMAN: OK. Flip can also be a biased coin. So for example, if I do-- I define something like, you know, define-- let's do this slightly differently. Let's call this lambda something. And what it does is flip 0.9. So if you run this forward, what you'll get now is that flip can actually take in some arguments. If you don't give it any arguments, it'll just do flip 50-50. If you give it some arguments, it'll do flip a biased coin, where the coin is biased towards 0.9. And you can see that, after I repeated that 1,000 times, I get, you know, it's approximately 90% heads, or true, and about 10% tails. AUDIENCE: Why did you make the lambda in there? TOMER ULLMAN: Ah, perfect, I'm glad somebody has asked that question. So if I were just to do the following-- suppose that I were just do repeat flip 0.9 like that, think about what would happen. What would happen is, I would first evaluate flip 0.9. OK, that would give me back a value, either true or false. And then this would say, repeat that 1,000 times. You would get, like, 1,000 trues, or 1,000 falses, or whatever it was that was first. In fact, it's going to fail, because repeat expects a function. But the point is, the reason that this is going to fail is because it wants a particular function. This is not a function, this is a value. You evaluate this first. It gives you a value like true or false. And then you repeat that value 1,000 times. That's not what you want. What you want is a procedure. A procedure, or a distribution, or something like that, some sort of function that, when you run it, you get a biased sample, so what would that look like? That would look like this. It would be-- or I could do something like this. Define my-coin weight-- OK, something like this. And what it does is this. Now what I've defined is, I've defined a procedure that takes in a particular weight. And what it does is that it gives you back a flip on that weight. AUDIENCE: [INAUDIBLE] TOMER ULLMAN: Yes, although you might, again, run into some problems, but we can get to that, because-- well, OK. So let's see-- AUDIENCE: How would define it as a lambda calculus? TOMER ULLMAN: OK, so how you would define it with the lambda calculus is, you would say my-coin lambda weight this thing. OK, now we're saying, what sort of thing is coin? Coin is a procedure. How do we know it's a procedure? Because we have this lambda right here. How many arguments does it expect? One, it's called weight. What does it do? It flips a coin. It gives you back that sample. AUDIENCE: Can I do-- TOMER ULLMAN: The equivalent way of doing that is by writing this thing without any lambdas. You would just write define my-coin-- notice the brackets there, right? Before we didn't have brackets around that-- define my-coin weight flip weight, like that. And now you're sort of saying, like, this is a procedure. You should know it's a procedure, because it's the first thing that you're hitting after define because of the parentheses. What sort of procedure is it? It's called my-coin. It takes in weight. Again, these are equivalent. And to answer Nori's question about how would I just do that without having to define things, I would say something like, hist repeat 1,000. Now, what do I want to repeat? I want to repeat some sort of procedure that samples things. So it's-- I'll call it lambda. It's an empty lambda. It doesn't take in any arguments. It's just the procedure. And what it does is, it flips a coin 0.9. And if I run that, I'll get that. OK, yes, no? OK, good. OK, so let's see, there are many other primitives that we could get to. There is uniform-draw. You can look at this online, but there's-- the basic primitives are things like multinomial, uniform, random integer, beta, Dirichlet, there's also the Chinese restaurant process. So let's see, we can build in our own little distribution. OK, let's try doing that. So here I've defined something which, under the hood, it's actually-- it's an interesting distribution. You all probably know it. But the way I'm going to define it is, I'm going to call it times it counts until heads. This is a procedure that's going to flip a coin. And if it comes up-- it's going to flip a coin with a particular weight. If it comes up true, if it comes up heads, then it's just going to stop. It's going to give you back 0. If it doesn't stop, if it comes back tails, it's going to tell you that. It's going to write down somewhere, like, 1. And it's going to keep going. It's going to recurse somehow, call itself, and then keep going. So this is for you, this is an exercise for you. You have it under the files, under 3.4, build your own distribution. I've left this open. Why don't you take two minutes. We're trying to build a procedure that gives me the amount of times that I need to flip a coin before I get back heads, OK? If I take a particular coin-- I guess I don't want to have one handy-- but I flip a coin. And I just-- you know, I flip it. If it comes back heads, I write down 0 and I'm done. If it comes back tails, I'm going to keep flipping it, so I flip it again. And you know, I might flip it 10 times until I get heads, so the point is that this procedure will, in that case, return 10. That would be one particular sample. Now, of course, if I take the coin again and I flip it again, sometimes I get 10 times until heads, sometimes once, sometimes 5, sometimes 20, so I'm going to get a particular distribution on the number of times I need until I hit heads. And the thing that we're trying to implement right now is just a procedure that, what it does is, it implements this counting thing that I just said by literally flipping a coin-- well, I don't know if literally, but under the hood, flipping a coin. If the coin comes back heads, because this thing evaluates to true, give back 0. If it doesn't, give back plus 1 plus what? So fill in those dots-- it shouldn't be a long expression-- such that you'll get what I was just talking about. So, guys, let me tell you what I was going for. An int plus 1 countsTillHeads coinweight. OK, and now if you do something like countsTillHeads, I don't know, 0.1 or something like that, and you run it. And it gets saved-- so let's read through this for a second. What happens is, you defined a procedure. It's called countsTillHeads. It takes in a coin weight. It flips a coin. If it comes back head, it gives you back 0. If it didn't come back heads, then you just do plus 1. And then you just call that thing again. You do countTillHeads coinweight again and again. If it comes back 0, then this time, you'll have plus 1 plus 0 if it came back heads in here. But if it didn't, then this will be plus 1 plus something. In effect, what we've defined here-- those of you that have defined it, and if not, just look at this-- what you've defined here is sort of a procedure that might give us back infinity in some way, except it's becoming extremely unlikely to do so with each particular flip of the coin. Now, I run it once with 0.1. I get 15. I can run it again and I'll get, you know, 8. That just means that, on that run, I flipped it eight times before I got heads. And again, I can do this many, many different times. Like, I can do hist repeat 1,000 and then this thing, some empty procedure that does that. And what you gets is this, which, in case it doesn't look familiar-- sorry, it's just the way these things usually look. This is sort of flipping the x- and y-axis. But the point is, how many times did I have to flip it to get, you know-- how many times did it happen? Did I flip it three times, or one, or two, three times? That's about 24%. And it sort of goes down, and down, and down, because it becomes much, much, much more unlikely that I'll flip it 40 times until I get heads. It could be that I'll keep flipping it to infinity, but it's not going to happen. This, in case you didn't know, falls off geometrically. It's the geometric distribution. That's a very fundamental, simple distribution. And one way to write it is to say, what's the probability of k? The probability of k is-- let's say, we have a coin which has the-- it's probability of coming up heads is p. Then we say the probability of k is p to the k minus 1 times 1 minus p, yes? It's I flip the coin 1 minus p times to the k. The point is, you can define the geometric distribution by sort of saying, what's the probability of any particular number? Or you can define the procedure for it, OK? Instead of writing down what should be the probability of any particular sequence, you can just write down the procedure that it describes. This is the procedure. The procedure doesn't explicitly tell you what the distribution is, it just samples it. You've built a procedure for flipping a coin. And if you do it many, many, many different times, what you'll get is the geometric distribution. This is will approach the geometric distribution. I can probably also do density, and then it'll show you it like that. So that's what I was talking about before with, like, trying to wrap your head around something like the equivalence between a probability distribution that you can write down in math or as an analytical expression and writing down the equivalent procedure for generating that probability distribution. Let's move on to something a little bit more interesting like Gaussian sampling. If you're not with us, you can look at it in 3.5, Gaussian Samples. What I've done here is, basically, I'm defining a particular center. Let's walk through this for a second. I'm defining a two-dimensional Gaussian. What it does is, it takes a particular center. A center is just an x-y point. And it does, you know, Gaussian around the first one. I'm trying to define a two-dimensional Gaussian. The way I do it is, I take a point around-- a one-dimensional Gaussian around this point. And I take a one-dimensional Gaussian around the second point. And then I just draw it. So in this particular case, I'm going to define my Gaussian center as 3, 2. OK, I'm going to take it x equals 3, y equals 2. And I want to sample a Gaussian around 3, 2. So I'm going to sample of Gaussian around 3 and a Gaussian around 2. And I'm going to give you that back. And if I repeat this 1,000 times, then-- and I scatter it, I'll end up with a plot that looks a bit like this. And you can see on the x-axis, this is 3. And this is 2. And it's basically a Gaussian with sampling points from around this thing, another forward procedure that I can sample. OK, is everyone more or less on board with this? Let's take two seconds to read this again. A basic procedure in Church is Gaussian. What I do is I basically-- I try to call Gaussian on some number. Gaussian takes in two arguments. Gaussian takes in a mean and a variance. In particular, I'm going to take a Gaussian. And its mean is going to be the first argument of center. Its variance it's going to be 1. I'm going to take a Gaussian sampled from the second argument, the y, and a variance of 1. And then I'm going to just give you back to that point. So this is a procedure that takes in a center point. And each time you sample it, it will give you a sample from around the mean 3, 2. And if I run that-- so now I've defined a particular center. You know, I've defined it 3, 2. I could have done many other different things. And I repeat that 100 times. I've basically drawn a sample from something around 3, 2. This can quickly get more interesting if you do something like a mixture of Gaussians. So a Gaussian mixture model is usually just saying, OK, I have some particular space. And I'm trying to figure out how many Gaussians are in this scene, so let's write down the forward model for that thing. What's the forward model for a mixture model? The forward model saying, I'm going to draw out some number of Gaussians. I don't know how many. And I don't necessarily know what their center point is, right? And from each one of these, I'm going to draw some number of samples. Does everyone understand, more or less, that description that I just gave? We're going to write it out now. But the point is, the generative model in your head for a mixture of Gaussians should be, there are some number of Gaussians. I don't know what it is. Each one of them is centered on some point. I don't know what it is. Let's say I know the variance just for simplicity, but I could obviously put a prior on that. And then I just sample from that. And I'll get some distribution. And then you could use-- we'll later on see, once you write down that forward model, it's pretty simple to then just invert it and say, OK, I see some number of points. How many Gaussians are there actually? But let's write down the forward model. So I have already done this ahead of time. And I'll do it here. So what I've done here, minus the typo, thanks, is to say something like, I want a sample of Gaussian center where I don't know where it is, but I'm going to say that it's in this two-dimensional space between 0 and 10, a box that's 10 wide and 10 tall. So for each new Gaussian, I don't know where its center is, but I'm assuming it's somewhere in this box that we're looking at. And the way I do that is, I say, OK, I define some sort of procedure. Each time you evaluate this procedure, what it's going to give you back is a pair, where the first thing in the pair is a uniform between 0 and 10, the second thing in the pair is a uniform between 0 and 10. If all you were to do are to sample Gaussian center, you would get back some number uniformly-distributed in the 10 box, where the first one is, let's say, x, and the second one is y. And the next thing I do is, let's say I want to define some number of Gaussians and I don't know how many there are. Let's say, for example, that I want to put some sort of ignorance prior on Gaussians between-- there might be one, there might be two, there might be 10. Let's say I stop it at 10 or something like that. So in this case, I just say, sample the number of Gaussians from something like random integer 10, since this goes to 0, and you don't want 0, I'm just adding the number 1 here. But what I also could have done, and I think I was going to do this is an exercise, but since we want to get to physics, and psychology, and some more interesting stuff, what I could have done here is define number of Gaussians-- suppose I wanted to put a prior on there being potentially an infinite number of Gaussian, what would I do? AUDIENCE: Dirichlet. TOMER ULLMAN: A Dirichlet, right? Or what else can I do that we've already learned? We could do the geometric, right? We just defined the geometric a second ago. The geometric gives us a probability on numbers basically going from 0 to infinity. And it dies off very quickly, so this gives us sort of a natural prior of some sort to say, I think that there are some number of Gaussians here. I don't know what it is. I'm pretty sure it dies off. Like, I don't think 100 is as equally likely as 10. I don't think 10 is as equally likely as 1. So I could have said, define number of Gaussians, just draw from geometric. And then I would have gotten some number, potentially infinite. You've just defined an infinite Gaussian mixture model. And then I draw some number of centers by basically repeating this procedure. I sample the Gaussians. And then I scatter the points. Let's see, and then you can look at the points. And this is a fun game to play. It's basically recapturing a bit of what Josh said before, which is to say, how many Gaussians do you think are in this image? And you can sort of play that with yourself to get a sense of it. You know, you've defined some procedure. You don't know how many Gaussians you actually created. You don't know exactly where they are, but you can run it forward. And you can look at it and say, well, here I think it's pretty obvious. I think there's sort of a Gaussian here, maybe a Gaussian here. So I guess the number here is 2, but here it's a bit less obvious. And again, you can play with this. So those of you who've written this down, and assuming you've done either a Dirichlet or a geometric distribution what you've basically done is written down the forward model for an infinite Gaussian mixture model. And you did it in, more or less, five lines of code. Yeah? AUDIENCE: What is the fold there? TOMER ULLMAN: Where do you see fold here? AUDIENCE: Visualize scatter fold append TOMER ULLMAN: Ah, yes, so fold is another high-level procedure. It's not terribly important for the purposes of this tutorial, but what it does is, it basically takes in a function. It takes in a list of stuff. And it basically applies it to the first argument. Then it takes it and applies it to whatever the result was plus the next item-- AUDIENCE: Plus? TOMER ULLMAN: --in the list. Well, not exactly plus-- AUDIENCE: In addition? TOMER ULLMAN: --but, yes, in addition, so you can have a fold which has, for example, two arguments. And what it does is it multiplies. So then you would take a list. And you would basically do-- or rather, what is sum. what some is basically is a fold of plus over a list, because it takes the first number, sums it up with the second one, takes that result, sums it up with a third one-- AUDIENCE: [INAUDIBLE] TOMER ULLMAN: Fold needs three arguments. Fold needs a particular-- well, it needs the function that you're going to apply. It needs a starting point to start from. And it needs a lot that it's going to work on, again, not terribly important for-- AUDIENCE: So why do this? TOMER ULLMAN: So in this particular case, what I'm trying to do in the background is, I'm going to get a lot of Gaussians. I don't know how many. I'm going to get basically a list of lists. It could be one. It could be three. It could be 10. Each one of them is going to define some number of points. And I just want to scatter them. But scatter works by taking in one list, so it's basically just a way of collapsing. Say I have three, or 10, I don't know how many. I'm trying to collapse some number of lists into a single list. We've defined some number of Gaussians. This is a London Blitz example. Josh was talking about this a little bit. Those of you who want to, sort of, jump back in again, you can go to 3.5.2 in the student document. You can copy and whatever is under that and paste it. And let's talk about that example for a second. What this thing is doing is, it's sort of Josh's example-- do you remember his example of, we have some sort of grid. And we're trying to say, is there a suspicious cluster somewhere, a disease cluster? We have some dots. And we're trying to figure out is there something going on here? You know, there's sort of a faulty, I don't know, whatever, asbestos or something like that. And I want to figure that out. So what you're going to get is sort of a 2D map. You're going to get some dots from that map. And you're trying to figure out-- your hypothesis is either this is sort of randomly-distributed, it's a uniform, or there's some sort of center here. So how do we write down the forward model for something like that? We would write down either-- the particular example, I'm doing here is another example that Tom Griffiths did, which is, during the Blitz, during the London bombing-- this is actually a very old example of finding patterns. Some of the British, the people of London, were convinced that there were spies in London that were telling the Germans where to bomb during the Blitz. And the way that they reasoned this is, they looked at the pattern of bombings. And they said, there's no way that this is random. They just looked at, like, dots on a map. And to them, it looked a bit like Gaussians, or things like that. They were working from, sort of, few examples. When you look at, there's, sort of, these nice web-- "nice," I don't know if it's nice-- but there's these websites that show you the entire Blitz from when it started to when it ended. And it's basically a random distribution. If you run statistical tests on it, it's no different from a random distribution. How would you run such a test on it? What you would do, for example, is you would write a forward model that says it's either random, uniform, or it's not. Now, tell me which one is more likely. And that's what people have, kind of, done. That's a nice data set to play around with. The way that we've written it over here is to say, look, we have two options. Either it's a uniform bombing or it's some targeted bombing. The uniform bombing is basically going to give us just some point between 0-- between this box of 0 to 10, just this thing that we were talking about before. It's going to sample uniformly from this box. The targeted bombing is going to sample some Gaussians, just like we defined before. You don't know how many. You don't know where the center is. And it's going to then sample from those Gaussians. And it's going to give you back some sort of scatter. And you're basically going to say, OK, I don't know if it's random, uniform, or if there's some targeted bombing going on here, so I'm going to place, basically, some inference. I'm going to flip a coin. If it comes up heads, I'm going to do uniform bombing. If it comes up tails, I'm going to do targeted bombing. And then you could look at something like this. And you can say, well, I don't know. That's kind of odd. I mean, it doesn't exactly look like a uniform bombing. There's all this missing empty space over here, right? It doesn't exactly look like one particular target. And again, you can sort of play with this. And we'll get into the inference about how to invert this thing. But just as a forward model, you can play with this, run it forward, and try to see if you can guess.
MIT_RES9003_Brains_Minds_and_Machines_Summer_Course_Summer_2015	Lecture_23_Josh_Tenenbaum_Computational_Cognitive_Science_Part_3.txt	The following content is provided under a Creative Commons license. Your support will help MIT OpenCourseWare continue to offer high quality educational resources for free. To make a donation or view additional materials from hundreds of MIT courses, visit MIT OpenCourseWare at ocw.mit.edu. JOSH TENENBAUM: So where we left off was, you know, again, I was telling you a story, both conceptual and motivational and a little bit technical about how we got to the things we're trying to do now as part of the center. And it involves, again both the problems we want to solve. We want to understand what is this common sense knowledge about the physical world and the psychological world that you can see in some form even in young infants? And what are the learning mechanisms that build it and grow it? And then what's the kind of technical ideas that are going to be also hopefully useful for building intelligent robots or other AI systems that can explain on the scientific side how this stuff works? All right, so that was all this business. And what I was suggesting, or I'll start to suggest here now, is that it goes back to that quote I gave at the beginning from Craik, right? The guy who in 1943 wrote this book called The Nature of Explanation. And he was saying that's the essence of intelligence is this ability to build models that allow you to explain the world, to then reason, simulate, plan, and so on. And I think we need tools for understanding how the brain is a modeling engine or an explaining engine. Or to get a little bit recursive about it, since what we're doing in science is also an explanatory activity, we need modeling engines in which we can build models of the brain as a modeling engine. And that's where the probabilistic programs are going to come in. So part of why I spent a while in the morning talking about these graphical models and the ways that we tried and, I think, made progress on, but ultimately we're dissatisfied with, talking about how we're modeling various aspects of cognition with these kinds of graphical models. I put up-- I didn't say too much about the technical details. That's fine. You can read a lot about it or not. But these ways of using graphs, mostly directed graphs, to capture something about the structure of the world. And then you put probabilities on it in some way, like a diffusion process or a noisy transmission process for a food web. That's a style of reasoning that sometimes goes by the name of Bayesian networks or causal graphical models. It's been hugely influential in computer science, many different fields, not just AI, and many fields outside of computer science. Not just cognitive science, neuroscience-- many areas of science and engineering. Here are just a few examples of some Bayesian networks you get if you search on Google image for Bayesian networks. And if you look carefully, you'll see they come from biology, economics, many chemical engineering, whatever. They're due to many people. Maybe more than anyone, the person who's most associated with this idea and with the name Bayesian networks is Judea Pearl. He received the Turing Award, which is like the highest reward in computer science. This is a language that we were using in all the projects you saw up until now in some form that we and many others used. Because they provide a powerful set of tools for general purpose tools. It goes back to this dream of building general purpose systems for understanding the world. So these provide general purpose languages for representing causal structure-- I'll say a little bit more about that-- and general purpose algorithms for doing the probabilistic inference on this. So we talked about ways of combining sophisticated statistical inference with knowledge representation that's causal and compositional. These models-- I'll just tell you a little bit about the one in the upper left up there, that thing that says diseases and symptoms. It is causal. It is compositional. It does support probabilistic inference. And it was the heart of why we were doing what we were doing and showing you how different kinds of causal graphical models basically could capture different modes of people's reasoning. And the idea that maybe learning about different domains was learning those different kinds of graphs structures. So let me say a little bit about how it works and then why it's not enough, because it really isn't enough. I mean, it's the right start. It's definitely in the right direction. But we need to go beyond it. That's where the problematic programs come in. So look at that network up there on the upper left. It's one of the most famous Bayesian networks. It's a textbook example. One of the first actually implemented AI systems was based on this for a system for medical diagnosis. Sort of a simple approximation to what a general practitioner might be doing if a patient comes in and reports some pattern of symptoms, and they want to figure out what's wrong. So diagnosis of a disease to explain the symptoms. The graph is a bipartite graph. So two sets of nodes with the arrows, again, going down in the causal direction. The bottom layer, the symptoms, are the things that you can nominally observe. A patient comes in reporting some symptoms. Not all are observed, but others maybe are things that you could test, like medical test results. And then the top level is this level of latent structure, the causes, the things that cause the symptoms. The arrows represent basically which diseases cause which symptoms. In this model there's roughly 500, 600 diseases-- you know, the commonish ones, not all that common-- and 4,000 symptoms. So it's a big model. And in some sense, you can think of it as a big probability model. It's a way of specifying a joint distribution on this 4,600-dimensional space. But it's a very particular one that's causally structured. It represents only the minimal causal dependencies and really only the minimal probabilistic dependencies. That sparsity is really important for how you use it, whether you're talking about inference or learning. So inference means observing patterns of symptoms or just observing the values of some of those variables and making guesses about the others. Like observing some symptoms and making guesses about the diseases that are most likely to have explained those. Or you might make a prediction about other symptoms you could observe. So you could go up and then back down. You could say, well, from these symptoms, I think the patient might have one of these two rare diseases. I don't know which one. But if it was this disease, then it would predict that symptom or that test maybe. But this disease wouldn't, so then that suggests a way to plan an action you could take to figure things out. So then I could go test for that symptom, and that would tell me which of these diseases the patient has. They're also useful in planning other kinds of treatments, interventions. Like if you want to cure someone-- again, we all know this intuitively-- you should try to cure the disease, not the symptom. If you have some way to act to change the state of one of those disease variables to kind of turn it off, reasonably that should relieve the symptoms. If that disease gets turned off, these symptoms should turn off. Whereas just treating the symptom like taking Advil for a headache is fine if that's all the problem is. But if it's being caused by something, you know, god forbid, like a brain tumor, it's not going to help. It's not going to cure the problem in the long term. OK, so all those patterns of causal inference, reasoning, prediction, action planning, exploration is a beautiful language for capturing all of those. You can automate all those inferences. Why isn't it enough, then, for capturing commonsense reasoning or this approach to cognition? Which I'm calling the kind of model-building explaining part, as opposed to the pattern recognition part. I mean, again, I don't want to get too far behind in talking about this. But that example is so rich. Like if you can build a neural network, you can just turn the arrows around to learn a mapping from symptoms to diseases, and that would be a pattern classifier. So these two different paradigms for intelligence-- as some of the questions we're getting at, and as I will show versions of that with some more interesting examples in a little bit-- often it's very subtle, and the relations between them are quite valuable. So one way to work with such a model, for example, or one nice way-- I mean, I mentioned a lot of people want to know-- and I'll keep talking about this for the rest of the hour-- productive ways to combine these powerful generative models with more pattern recognition approaches. For some classes of this model-- there are always general purpose algorithms that can support these inferences, that can tell you what diseases you're likely to have given what symptoms. But in some cases, they could be very fast. In other cases, they could be very slow. Whereas if you could imagine trying to learn a neural network that looks just like that, only the arrows go up, so they implement a mapping from data to diseases, that could help to do much faster inference in the cases where that's possible. So that's just one example of where a model, which might be not a crazy way to think about, for example, more generally the way top-down and bottom-up connections work in the brain. I'll take that a little bit more literally in a vision example in a second. So there's a lot you can get from studying these causal graphical models, including some version of what it is for the mind to explain the world and how that explanation and pattern recognition approach can work together. But it's not enough to really get at the heart of common sense The mental generative models we build are more richly structured. They're more like programs. What do I mean by that? Well, here I'm giving a bunch of examples of scientific theories or models. Not commonsense ones, but I think the same idea applies. Ways of, again, explaining the world, not just describing the pattern. So we went at the beginning through Newton's laws versus Kepler's laws. That's just one example. And you might not have thought of those laws as a program, but they're certainly not a graph. On the first slide when I showed Newton's laws, there was a bunch of symbols, statements in English, some math. But what it comes down to is basically a set of pieces of code that you could run to generate the orbits. It doesn't describe the sheep or the velocities, but it's a machine that you plug in some things. You plug in some masses, some objects, some initial conditions. And you press run, and it generates the orbits, just like what you're seeing there. Although those probably weren't generated. That's a GIF. OK. That's more like Kepler or Ptolemy. But anyway, it's a powerful machine. It's a machine, which if you put down the right masses in the right position, they don't just all go around in ellipses. Some of them are like moons, and they will go around the things that will go around the others. And some of them will be like apples on the Earth, and they won't go around anything. They'll just fall down. So that's the powerful machine. And in the really simplest cases, that machine-- those equations can be solved analytically. You can use calculus or other methods of analysis like Newton did. He didn't have a computer. And you can show that for a two-body system, one planet and one sun, you can solve those equations to show that you get Kepler's law. Amazing. And under the approximation that only the sun is-- for every other planet, it's only the sun that's exerting a significant influence, you can describe all of Kepler's laws this way. But once you have more than two bodies interacting in some complex way, like three masses similar in size near each other, you can't solve the equations analytically anymore. You basically just have to run a simulation. For the most part, the world is complicated, and our models have to be run. Here's a model of a riverbed formation. Or these are snapshots of a model of galaxy collision, you know, and climate modeling or aerodynamics. So basically what most modern science is is you write down descriptions of the causal processes, something going on in the world, and you study that through some combination of analysis and simulation to see what would happen. If you want to estimate parameters, you try out some guesses of the parameters. And you run this thing, and you see if its behavior looks like the data you observe. If you are trying to decide between two different models, you simulate each of them, and you see which one looks more like the data you observe. If you think there's something wrong with your model-- it doesn't quite look like the data you observe. You think, how could I change my model, which basically if I run it, it'll look more like the data I observe in some important way? Those activities of science-- those are, in some form I'm arguing, the activities of common sense explanation. So when I'm talking about the child as scientist, that's what I'm basically talking about. It's some version of that. And so that includes both using-- describing the causal processes with a program that you run. Or if you want to talk about learning, the scientific analog is building one of these theories. You don't build a theory, whether it's Newton's laws or Mendel's laws or any of these things, by just finding patterns and data. You do something like this program thing, but kind of recursively. Think of you having some kind of paradigm, some program that generates programs, and you use it to try to somehow search the space of programs to come up with a program that fits your data well. OK, so that's, again, kind of the big picture. And now, let's talk about how we can actually do something with this idea-- use these programs. And you might be wondering, OK, maybe I understand-- I'm realizing I didn't say the main thing I want you to understand. The main thing I want you to get from this is how programs go beyond graphs. So none of these processes here can nicely describe with a graph the way we have in the language of graphical models. So the interesting causality-- I mean, in some sense, there's kind of a graph. You can talk about the state of the world at time T, and I'll show you graphs like this in a second. The state of the world at time T plus 1 and an arrow forward in time. But all the interesting stuff that science really gains power from are the much more fine-grained structure captured in equations or functions that describe exactly how all this stuff works. And it needs languages like math or C++ or LISP. It needs a symbolic language of processes to really do justice to. The second thing I want to get, which will take a minute to get, but let's put it out there. As yes, OK, maybe you get the idea that programs can be used to describe causal processes in interesting ways. But where is the probability part come in? So the same thing is actually true in graphical models. How many people have read Judea Pearl's 2000 book called Causality? How many people have read his '88 book? Or nobody's read anything. But, OK, so what Pearl is most famous for-- I mean, when we say Pearl's famous for inventing Bayesian networks, that's based on work he did in the '80s, in which, yes, they were all probability models. But then he came to what he calls, and I would call, too, a deeper view in which it was really about basically deterministic causal relations. Basically it was a graphical language for equations-- certain classes of equations like structural equations. If you know about linear structural equations, it was sort of like nonlinear structural equations. And then probabilities are these things you put on just on top of it to capture the things you don't know that you're uncertain about. And I think he was he was getting at the fact that to scientists, and also to people-- there's some very nice work by Laura Schultz and Jessica Sommerville, both of whom will be here next week actually, on how children's concepts of causality are basically deterministic at the core. And where the probabilities come in is on the things that we don't observe or the things we don't know, the uncertainty. It's not that the world is noisy. It's that we believe, at least-- except for quantum mechanics-- but our intuitive notions are that the world is basically deterministic, but with a lot of stuff we don't know. This was, for example, Laplace's view in philosophy of science. And really until quantum mechanics, it was broadly the Enlightenment science view that the world is full of all these complicated deterministic machines, and where uncertainty comes from the things that we can't observe or that we can't measure finely enough, or they're just in some form unknown or unknowable to us. Does that make sense? So you'll see more of this in a second. But where the probabilities are going to come from is basically if there are inputs to the program that we don't know or parameters we don't know that in order to simulate them we're going to have to put distributions on those and make some guesses and then see what happens for different guesses. Does that make sense? OK. Good. So again, that's most of the technical stuff I need to say. And you'll learn about how this works in much more concrete details if you go to the tutorial afterwards that Tomer is going to run. What you'll see there is this. So here are just a few examples. Many of you hopefully already looked at the web pages from this probmods.org thing. And what you see here is basically each of these boxes is a probabilistic program model. Most of it is a bunch of defined statements. So if you look here, you'll see these defined statements. Those are just defining functions. They name the function. They take some inputs, which call other functions, and then they maybe do something-- they have some output that might be an object. It might itself be a function. These can be functions that generate other functions. And where the probabilities come in is that sometimes these functions call random number of generators basically. If you look carefully, you'll see things like Dirichlet, or uniform draw, or Gaussian, or flip. Right those are primitive random functions that flip a coin, or roll a die, or draw from a Gaussian. And those captured things that are currently unknown. In a very important sense, the particular language, Church, that you're going to learn here with its sort of stochastic LISP-- basically just functions that call other functions and maybe add in some randomness to that-- is very much analogous to the directed graph of a Bayesian network. In a Bayesian network, you have nodes and arrows. And the parents of a node, the ones that send arrows to it, are basically the minimal set of variables that if you were going to sample from this model you'd have to sample first in order to then sample the child variable. Because those are the key things it depends on. And you can have a multi-layered Bayesian network that, if you are going to sample from it, it's just you start at the top and you sort of go down. That's exactly the same thing you have in these probabilistic programs where the defined statements are basically defining a function. And the functions are the nodes, and the other functions that they call as part of the statement are the things that are the nodes that send arrows there. But the key is, as you can imagine if you've ever-- I mean, all of you have written computer programs-- is that only very simple programs look like directed acyclic graphs. And that's what a Bayesian network is. It's very easy and often necessary to write a program to really capture something causally interesting in the world where it's not a directed acyclic route. There's all sorts of cycles. There's recursion. One thing that a function can do is make a whole other graph. Or often it might be directed and acyclic, but all the interesting stuff is kind of going on inside what happens when you evaluate one function. So if you were to draw it as a graph, it might look like you could draw a directed acyclic graph, but all the interesting stuff will be going on inside one node or one arrow. So let me get more specific about the particular kind of programs that we're going to be talking about. In a probabilistic programming language like Church, or in general in this view of the mind, we're interested in being able to build really any kind of thing. Again, there's lots of big dreams here. Like I was saying before, I felt like we had to give up on some dreams, but we've replaced it with even grander ones, like probabilistic modeling engines that can do any computable model. But in the spirit of trying to scale up from something that we can get traction on, what I've been focusing on in a lot of my work recently and what we've been doing as part of the center, are particular probabilistic programs that we think can capture this very early core of common sense intuitive physics and intuitive psychology in young kids. It's what I called-- and I remember I mentioned this in the first lecture-- this game engine in your head. So it's programs for graphics engines, physics engines, planning engines, the basic kinds of things you might use to build one of these immersive video games. And we think if you wrap those inside this framework for probabilistic inference, then that's a powerful way to do the kind of common sense seen understanding, whether in these adult versions or in the young kid versions. Now, to specify this probabilistic programs view, just like with Bayesian networks or these graphical models, we wanted general purpose tools for representing interesting things in the world and for computing the inferences that we want. Again, which means basically observing, say, just like you observe some of the symptoms and you want to compute the likely diseases that best explain the observed symptoms. Here we talk about observing the outputs of some of these programs, like the image that's the output of a graphics program. And we want to work backwards and make a guess at the world state, the input to the graphics engine that's most likely to have produced the image. That's the analog of getting diseases from symptoms. Or again, that's our explanation right there. And there are lots of different algorithms for doing this. I'm not going to say too much about them. Tomer will say a little bit more in the afternoon. The main thing I will do is, I will say that the main general purpose algorithms for inference in probabilistic programming language are in the category of slow and slower and really, really slow. And this is one of the many ways in which there's no magic or no free lunch. Across all of AI and cognitive science, when you build very powerful representations, doing inference with them becomes very hard. It's part of why people often like things like neural networks. They're much weaker representations, but inference can be much faster. And at the moment, the only totally general purpose algorithms for doing inference with probabilistic programs are slow. But first of all, they're getting faster. People are coming up with-- and I can talk about this offline where that's going-- but also-- and this is what I'll talk about in a sharper way in a second-- there are particular classes of probabilistic programs, in particular, the ones in the game engine in your head. Like for vision is inverse graphics and maybe some things about physics and psychology too. I mean, again, I'm just thinking of the stuff of like what's going on with it when a kid is playing with some objects around them and thinking about what other people might think about those things. It's just that setting where we think that you can build sort of in some sense special purpose. I mean, they're still pretty general. But inference algorithms for doing inference in probabilistic programs, getting the causes from the effects that are much, much faster than things that could work on just arbitrary probabilistic programs and that actually often look a lot like neural networks. And in particular, we can directly use, say for example, deep convolutional neural networks to build these recognition programs or basically inference programs that work by pattern recognition in, for example, an inverse graphics approach to vision. So that's what I'll show you basically now. I'm going to start off by just working through a couple of these arrows. I'm going to first talk about this sort of approach we've done to tackle both vision as inverse graphics and some intuitive physics on the scene recovered and then say a little bit about the intuitive psychology side. Here's an example of the kind of specific domain we've studied. It's like our Atari setting. It's a kind of video game inspired by the real game Jenga. Jenga's this cool game you play with wooden blocks. You start off with a very, very, very nicely stacked up thing and you take turns removing the blocks. And the player who removes the block that makes the whole thing fall over is the one who loses. And it really exercises this part of your brain that we've been studying here, which is an ability to reason about stability and support I very briefly went over this, but this is something that is one of the classic case studies of infant object knowledge. Looking at how basically these concepts develop in some really interesting ways over the first year of life. Though what we're doing here is building models and testing them primarily with adults. It is part of what we're trying to do in our brains, minds, and machines research program here, collaboration with Liz and others, to actually test these ideas in experiments with infants. But what I'll show you is just kind of think of it as like infant-inspired adult intuitive physics where we build and test the models in an easier way, and then we're taking it down to kids going forward. So the kind of experiment we can do with adults is show them these configurations of blocks and say, for example, how stable under gravity is one of these towers or configurations? So like everything else, you can make a judgment on a scale of zero to 10 or one to seven. And probably most people would agree that the ones in the upper left are relatively stable, meaning if you just sort of run gravity on it it's not going to fall over. Whereas the ones in the lower right are much more likely to fall under gravity. Fair enough? That's what people say. OK. So that's the kind of thing we'd like to be able to explain as well as many other judgments you could make about this simple, but not that simple world of objects. And again, you can see how in principle this could very nicely interface with what Demis was talking about. He talked about their ambition to do the SHRDLU task, which was this ability to basically have a system that can take in instructions and language and manipulate objects and blocks world. They are very far from that. Everybody's really far from having a general purpose system that can do that in any way like a human does. But we think we're building some of the common sense knowledge about the physical world that would be necessary to get something like that to work or to explain how kids play with blocks, play with each other, talk to each other while they're playing with blocks and so on. So the first step is the vision part. In this picture here, it's that blue graphics arrow. Here's another way into it. We want to be able to take a 2D image and work backwards to the world state, the kind of world state that can support physical reasoning. Again, remember these buzzwords-- explaining the mind with generative models that are causal and compositional. We want a description of the world which supports causal reasoning of the sort that physics is doing, like forces interacting with each other. So it's got to have things that can exert force and can suffer forces. It's got to have mass in some form. It's got to be compositional because you've got to be able to pick up a block and take it away. Or if I have these blocks over here and these blocks over here and I want to put these ones on top of there, the world state has to be able to support any number of objects in any configuration and to literally compose a representation of a world of objects that are composed together to make bigger things. So really the only way we know how to do that is something like what's sometimes in engineering called a CAD model or computer-aided design. But it's basically a representation of three-dimensional objects, often with something like a mesh or a grid of key points with their masses and springs for stiffness, something like that. Here my only picture of the world state looks an awful lot like the image, only it's in black and white instead of color. But the difference is that the thing on the bottom is actually an image. Whereas the thing on the top is just a 2D projection of a 3D model. I'll show you that one. Here's a few others. So I'll go back and forth between these. Notice how it kind of looks like the blocks are moving around. So what's actually going on is these are samples from the Bayesian posterior in an inverse graphics system. We put a prior on world states, which is basically a prior on what we think the world is made out of. We think there's these Jenga blocks basically. And then the likelihood, which is that that forward model is the probability of seeing a particular 2D image given a 3D configuration of blocks. And going back to the thing you had, it's basically deterministic with a little bit of noise. It's deterministic. It just follows the rules of OpenGL graphics. Basically says objects have surfaces. They're not transparent. You can't see through them. That's an extra complication if you wanted to have that. And basically the image is formed by taking the closest surface of the closest object and bouncing a ray of light off of it, which really just means taking its color and scaling it by intensity. It's a very simple shadow model. So that's the causal model. And then we can add a little bit of uncertainty like, for example, maybe we can't-- there's a little bit of noise in the sensor data. So you can be uncertain about exactly the low level image features. And then when you run one of these probabilistic programs in reverse to make a guess of what configuration of blocks is most likely to have produced that image, there is a little bit of posterior uncertainty that inherits from the fact that you can't perfectly localize those objects in the world. So again, what you see here are three or four samples from the posterior-- the distribution over best guesses of the world state of 3D objects that were most likely to have rendered into that 2D image. And any one of those is now an actionable representation for physical manipulation or reasoning. OK? And how we actually compute that, again, I'm not going to go into right now. I'll go into something like it in a minute. But at least in its most basic form, it involves some rather unfortunately slow random search process through the space of blocks models. Here's another example. This is another configuration there-- another image. And here is a few samples again from the posterior. And hopefully when you see these things moving around, whether it's this one or the one before, you see them move a little bit, but most of them look very similar. You'd be hard pressed to tell the difference if you looked away for a second between any one of those. Which one are you actually seeing? And that's exactly the point. The uncertainty you see there is meant to capture basically the uncertainty you have in a single glance at an image like that. You can't perfectly tell where the blocks are. So basically any one of these configurations up here is about equally good. And we think your intuitive physics, your sort of common sense core intuitive physics that even babies have, is operating over one or a few samples like that. Now in separate work that is not really-- I think of it as really about common sense, but it's one of the things we've been doing in our group and in CBMM where are these ideas best make contact with the rest of what people are doing here and where we can really test interesting neural hypotheses potentially and understand the interplay between these generative models for explanation and the more sort of neural-network-type models for pattern recognition. We've been really pushing on this idea vision as inverse graphics. So I'll tell you a little bit about that because it's quite interesting for CBMM. But I want to make sure to only do this for about five minutes and then go back to the how this gets used for more the intuitive physics and planning stuff. So this is this an example from a paper by Tejas Kulkarni who's one of our grad students. And it's joint work with a few other really smart people such as Vikash Mansinghka who's a research scientist at MIT and Pushmeet Kohli who's at Microsoft Research. And it was a computer vision paper, pure computer vision paper from the summer, where he was developing a specific kind of probabilistic programming language, but a general one for doing this kind of vision as inverse graphics where you could give a number of different models. Here I'll show you one for faces, another one for bodies, another one for generic objects. But basically you can pretty easily specify a graphics model that when you run it in the forward direction generates random images of objects in a certain class. And then you can run it in the reverse direction to do scene parsing to go from the image to the underlying scene. So here's an example of this in faces where the graphics model-- it's really very directly based on work that Thomas Vetter, who was a former student or post-doc of Tommy's actually, so kind of a early ancestor of CBMM, built. And his group in Basel, Switzerland, where it's a simple but still pretty nice graphics model for making face images. There's a model of the shape of the face, which again, it's like a CAD model. It's a mesh surface description. Pretty fine-grained structure of the 2D surface of the face in 3D. And there is about 400 dimensions to characterize the possible shapes of faces. And there's another 400 dimensions to characterize the texture, which is like the skin, the beard, the eyes, the color, and surface properties that get mapped on top of the mesh. And then there's a little bit more graphic stuff, which is generic, not specific to faces. That stuff is all specific to faces. But then there is a simple lighting model. So you basically have a point light source somewhere out there and you shine the light on the face. It can produce shadows, of course, but not very complicated ones. And then there's a viewpoint camera thing. So you put the light source somewhere and you put a camera somewhere specifying the viewpoint. And the combination of these, shape, texture, lighting, and camera, give you a complete graphic specification. It produces an image of a particular face lit from a particular direction and viewed from some particular viewpoint and distance. And what you see on the right are random samples from this probabilistic program, this generative model. So you can just write this program and press Go, Go, Go, Go, Go, and every time you run it, you get a new face viewed from a new direction and lighting condition. So that's the prior. Now, what about inference? Well, the idea of vision as inverse graphics is to say take a real image of a face like that one and see if you can produce from your graphics model something that looks like that. So, for example, here in the lower left is an example of a face that was produced from the graphics model that hopefully most of you agree looks kind of like that. Maybe not exactly the same, but kind of enough. And in building this system-- this system, by the way, is called Picture. That's that first word of the paper title, too, the Kulkarni, et al. paper. There were a few neat things that had to be done. One of the things that had to be done was to come up with various ways to say what does it mean for the output of the graphics engine to look like the image. In the case of faces, actually matching up pixels is not completely crazy. But for most vision problems, it's going to be unrealistic and unnecessary to build a graphics engine that's pixel-level realistic. And so you might, for example, want to have something where the graphics engine hypothesis is matched to the image with something like some kind of features. Like it could be convolutional neural network features. That's one way to use, for example, neural networks to make something like this work well. And Jojen just showed me a paper by some other folks from Darmstadt, which is doing what looks like a very interesting similar kind of thing. Let me show what inference looks like in this model and then say what I think is an even more interesting way to use convolutional. And that's from another recent paper we've been looking at. So here is, if you watch this, this is one observed face. And what you're seeing over here is just a trace of the system kind of searching through the space of traces of the graphics program. Basically trying out random faces that might look like that face there. It's using a kind of MCMC inference. It's very similar to what you're going to see from Tomer in the tutorial. It basically starts off with a random face and takes a bunch of small random steps that are biased towards making the image look more and more like the actual observed image. And at the end, you have something which looks almost identical to the observed face. The key, right, though, is that though the observed face is literally just a 2D image, the thing you're seeing on the right is a projection of a 3D model of a face. And it's one that supports a lot of causal action. So here just to show you on a more interesting sort of high-resolution set of face images, the ones on the left are observed images. And then we fit this model. And then we can rotate it around and change the lighting. If we had parameters that control the expression-- there's no real expression parameters here-- that wouldn't be too hard to put in. You could make us happy or sad. But you can see-- hopefully what you can see is that the recovered model supports fairly reasonable generalization to other viewpoints and lighting conditions. It's the sort of thing that should make for more robust face recognition. Although that's not the main focus of what we're trying to use it here. I just want to emphasize there's all sorts of things that would be useful if you had an actual 3D model of the face you could get from a single image. Or here's the same kind of idea now for a body pose system. So now, the image we're going to assume has a person in it somewhere doing something. Remember back to that challenge I gave at the beginning about finding the bodies in a complex scene like the airplane full of computer vision researchers where you found the right hand or the left toe. So in order to do that, we think you have to have something like an actual 3D model of a body. What you see on the lower left is a bunch of samples from this. So we basically just took a kind of interesting 3D stick figure skeleton model and just put some knobs on it. You can tweak it around. You can put some simple probability models to get a prior. And these are just random samples of random body positions. And the idea of the system is to kind of search through that space of body positions until you find one, which then when you project it from a certain camera angle looks like the body you're seeing. So here is an example of this in action. This is some guy-- I guess Usain Bolt. Some kind of interesting slightly unusual pose as he's about to break the finish line maybe. And here is the system in action. So it starts off from a random position and, again, sort of takes some takes a bunch of random steps moving around in 3D space until it finds a configuration, which when you project it into the image looks like what you see there. Now, notice a key difference when I say looks like-- it doesn't look like it at the pixel level like the face did. It's only matching at the level of these basically enhanced edge statistics which you see here. So this is an example of building a model that's not a photorealistic render. The graphics model is not trying to match the image. It's trying to match this. Or it could be, for example, some intermediate level of convonet features. And we think this is very powerful. Because more generally while we might have a really detailed model of facial appearance, for bodies, we don't have a good clothing model. We're not trying to model the skin. We're just trying to model just enough to solve the problem we're interested in. And again, this is reflective of a much more broad theme in this idea of intelligence as explanation, modeling the causal structure of the world. We don't expect, even in science, but certainly not in our intuitive theories, to model the causal structure of the world at full detail. And a way that either I am always misunderstood or always fail to communicate-- it's my fault really-- is I say, oh, we have these rich models of the world. People often think that means that somehow the complete thing. Like if I say we have a physics engine in our head, it means we have all of physics. Of if I say we have a graphics engine, we have all of every possible thing. This isn't Pixar. We're not trying to make a beautiful movie, except maybe for faces. We're just trying to capture just the key parts, just the key causal parts of the way things move in the world as physical objects and the way images are formed that at the right level of abstraction that matters for us allows us to do what we need to do. This is just an example of our system solving some pretty challenging body pose recognition problems in 3D, cases which are problematic even for the best of standard computer vision systems. Either because it's a weird pose, like these weird sports figures, or because the body is heavily occluded. But I think, again, these are problems which people solve effortlessly. And I think something like this is on the track of what we want to do. You can apply the same kind of thing to more generic objects like this, but I'm not going to go into the details. The last thing I want to say about vision before getting back to common sense for a few minutes-- and in some sense, maybe this is the most important slide for the broader CBMM, brains, minds, and machines thing. Because this is the clearest thing I can point to to the thing I've been saying all along since the beginning of the morning about how we want to look for ways to combine the generative model view and the pattern recognition view. So the generative model is what you see on the left here. It's the arrows going down. It's exactly just the face graphics engine, the same thing I showed you. The thing on the right with the arrows going up is a convonet. Basically it's a out-of-the-box, cafe-style, convolutional neural net with some fully connected layers on the top. And then there's a few other dashed arrows which represent linear decoders from layers of that model to other things, which are basically parts of the generative model. And the idea here-- this is work due to Ilker Yildirim, who some of you might have met. He was here the other day. He's one of our CBMM postdocs, but also joint with Tejas and with Winrich who you saw before. It's to try to in several senses combine the best of these perspectives, to say, look, if we want to recognize anything or perceive the structure of the world richly, I think it needs to be something like this inverse graphics or inverting a graphics program. But you saw how slow it was. You saw how it took a couple of seconds at least on our computer just for faces to search through the space of faces to come up with a convincing hypothesis. That's way too slow. It doesn't take that long. We know a lot about exactly how long it takes you from Winrich, and Nancy's, and many other people's work. So how can vision in this case, or really much more generally, be so rich in terms of the model it builds, yet so fast? Well, here's a proposal, which is to take the things that are good at being fast like the pattern recognizers, deep ones, and train them to solve the hard inference problem or at least to do most of the work. It's an idea which is very heavily inspired by an older idea of Geoff Hinton's sometimes called the Helmholtz machine. Here the idea in common with Hinton is to have a generative model and a recognition model where the recognition model is a neural network and it's trained to invert the generative the model. Namely, it's trained to map not from sense data to task output, but from sense data to the hidden deep causes of the generative model, which then, when you want to use this to act to plan what you're going to do, you plan on the model. To make an analogy to say the DeepMind video game player, this would be like having a system which, in contrast to the DeepQ network, which mapped from pixel images to joystick commands, this would be like learning a network that maps from pixel images to the game state, to the objects, the sprites that are moving around, the score, and so on, and then plans on that. And I think that's much more like what people do is that. Here just in the limited case of faces, what are we doing, right? So what we've got here is we take this convolutional neural network. We train it in ways that you can read about in the paper. It's very easy kind of training to basically make predictions, to make guesses about all the latent variables, the shape, the texture, the lighting, the camera angle. And then you take those guesses, and they start off that Markov chain. So instead of starting off at a random graphics hypothesis, you start off at a pretty good one and then refine it a little bit. What you can see here in these blue and red curves is the blue curve is the course of inference for the model I showed you before, where you start off at a random guess, and after, I don't know, 100 iterations of MCMC, you improve and you kind of get there. Whereas the red curve is what you see if you start off with the guess of this recognition model. And you can see that you start off sort of in some sense almost as good as you're ever going to get, and then you refine it. Well, it might look like you we're just were refining it a little bit. But this is a kind of a double log scale. It's a log plot of log probability. So what looks like a little bit there on the red curve is actually a lot-- I mean perceptually. You can see it here where if you take-- on the top I'm showing observed input faces. On the bottom I'm showing the result of this full inverse graphics thing. And they should look almost identical. So the full model is able to basically perfectly invert this and come up with a face that really does look like the one on the top. The ones in the middle are the best guess you get from this neural network that's been trained to approximately invert the generative model. And what you can see is on first glance it should look pretty good. But if you pay a little bit of attention, you can see differences. Like hopefully you can see this person is not actually that person in a way that this is much more convincingly. Or this person-- this one is pretty good, but I think this one-- I think it's pretty easy to say, yeah, this isn't quite the same person as that one. Do you guys agree? We've done some experiments to verify this. But hopefully they should look pretty similar, and that's the point. How do you combine the best of these computational paradigms? How can perception more generally be so rich and so fast? Well, quite possibly like this. It even actually might provide some insight into the neural circuitry that Winrich and Doris Tsao and others have mapped out. We think that this recognition model that's trained to invert the graphics model can provide a really nice account of some of Winrich's data like you saw before. But I will not go into the details because in maybe five to 10 minutes I want to get back to physics and psychology. So physics-- and there won't be any more neural networks. Because that's about as much-- I mean, I think we'd like to take those ways of integrating the best of these approaches and apply them to these more general cases. But that's about as far as we can get. Here what I want to just give you a taste of at least is how we're using ideas just purely from probabilistic programs to capture more of this common sense physics and psychology. So let's say we can solve this problem by making a good guess of the 3D world state from the image very quickly inverting this graphics engine. Now, we can start to do some physical reasoning, a la Craik's mental model of in the head of the physical world, where we now take a physics engine, which is-- here again we're using the kind of physics engines that game physics-- like very simple-- again, I don't have time to go into the details. Although Tomer has written a very nice paper with, well, with himself. But he's nicely put my name and Liz's on it-- about sort of trying to introduce some of the basic game engine concepts to cognitive scientists. So hopefully we'll be able to show you that soon too. Or you can read about them. Basically it's that these physics engines are just doing again a very quick, fast, approximate implementation of certain aspects of Newtonian mechanics. Sufficient that if you run it a few times steps with a configuration of objects like that you might get something like what you see over there on the right. That's an example of running this approximate Newtonian physics forward a few time steps. Here's another sample from this model, another kind of mental stimulation. We take a slightly different guess of the world state, and we run that forward a few time steps, and you see something else happens. Nothing here is claimed to be accurate in the ground truth way. Neither one of these is exactly the right configuration of blocks. And you run this thing forward, and it only approximately captures the way blocks really bounce off each other. It's a hard problem to actually totally realistically simulate. But our point is that you don't really have to. You just have to make a reasonable guess of the position of the blocks and a reasonable guess of what's going to happen a few time steps in the future to predict what you need to know and common sense, which is that, wow, that's going to fall over. I better do something about it. And that's what our experiment taps into. We give people a whole bunch of stimuli like the ones I showed you and ask them, on some graded scale, how likely do you think it is to fall over? And what you see here-- this is again one of those plots that always are the same where on the y-axis are the average human judgments now of-- it's an estimate of how unstable the tower is. It's both the probability that it will fall, but also how much of the tower will fall. So it's like the expected proportion of the tower that's going to fall over under gravity. And along the x-axis is the model prediction, which is just the average of a few samples from what I showed you. You just take a few guesses of the world state, run it forward a few time steps, count up the proportion of blocks it fell, and average that. And what you can see is that does a really nice job of predicting people's stability intuitions. I'll just point to an interesting comparison. Because it does come into where. Where does the probability come in in these probabilistic programs? Well, here's one very noticeable way. So if you look down there on the lower right, you'll see a smaller version of a similar plot. It's plotting now the results of-- it says ground truth physics, but that's a little misleading maybe. It's just a noiseless physics engine. So we take the same physics model, but we get rid of any of the state uncertainties. So we tell it the true position of the blocks, and we give it the true physics. Whereas our probabilistic physics engine allows for some uncertainty in exactly which forces are doing what. But here we say we're just going to model gravity, friction, collisions as best we can. And we're going to get the state of the blocks perfectly. And because it's noiseless, you notice that-- so those crosses over there are crosses because they're arrow bars, both across people and model simulations. Now they're just vertical lines. There's no arrow bars in the model simulation because it's deterministic. It's graded because there's the proportion of the tower that falls over. But what you see is the model is a lot worse. It scatters much more. The correlation dropped from around 0.9 to around 0.6 in terms of correlation of model with people's judgments. And you have some cases like this red dot here-- that corresponds to this stimulus-- which goes from being a really nice model fit. This is one which people judged to be very unstable, and so does the probabilistic physics engine. But actually it's not unstable at all. It's actually perfectly stable. The blocks are actually just perfectly balanced so that it doesn't fall. Although I'm sure everybody looks at that and finds that hard to believe. So this is nice. This is a kind of physics illusion. There are real world versions of this out on the beaches not too far from here. It's a fun thing to do to stack up objects in ways that are surprisingly stable. We would say a surprise because your intuitive physics has certain irreducible noise. What we're suggesting here is that your physical intuitions-- you're always in some sense making a guess that's sensitive to the uncertainty about where things might be and what forces might be active on the world. And it's very hard to see these as deterministic physics, even when you know that that's exactly what's going on and that it is stable. Let me say just a little bit about planning. So how might you use this kind of model to build some model of this core intuitive psychology? And I don't mean here all of theory of mind. Next week, we'll hear a lot more. Like Rebecca Saxe will be down here. We'll hear a lot more about much richer kinds of reasoning about other people's mental states that adults and older children can do. But here we're talking about, just as we were talking about what I was calling core intuitive physics, again inspired by Liz's work of just you know what objects do right here on the table top around us over short time scales, the core theory of mind, something that even very young babies can do in some form, or at least young children. There's controversy over exactly what age kids can be able to do this sort of thing. But in some form I think before language, it's the kind of thing that when you're starting to learn verbs, the earliest language is kind of mentalistic and builds on this knowledge. And take the red and blue ball chasing scene that you saw, remember from Tomer. That was 13-month-olds. So there's definitely some form of kind of interpretation of beliefs and desires in some protoform that you can see even in infants of around one year of age. And it's exactly that kind of thing also. Remember that, if you saw John Leonard's talk yesterday-- he was the robotics guy who talked about self-driving cars and how there's certain gaps in what they can do despite the all the publicity, like the can't turn left basically in an unrestricted intersection. Because there's a certain kind of theory of mind in street scenes when cars could be coming and people could be crossing or all those things about the police officers. Part of why this is so exciting to me and why I love that talk is because this is, I think, that same common sense knowledge that if we can really figure out how to capture this reasoning about beliefs and desires in the limited context where desires are people moving around in space around us and the beliefs are who can see who and who can see who can see who-- in driving, the art of making eye contact with other drivers or pedestrians is seeing that they can see you or that they can see what you can see and that they can see you seeing you them. It doesn't have to be super deeply recursive, but it's a couple of layers deep. We don't have to think about it consciously, but we have to be able to do it. So that's the kind of core belief desire theory of mind reasoning. And here's how we've tried to capture this with probabilistic programs. This is work that Chris Baker started doing a few years ago. And a lot of it joint Rebecca Saxe and also some of it with Julian Jara-Ettinger and some of it with Tomer. So there's a whole bunch of us who've been working on versions of this, but I'll just show you one or two examples. Again, the key programs here are not graphics or physics engines, but planning engines and perception engines. So very simple kinds of robotics programs, far too simple in this form to build a self-driving car or a humanoid robot, but maybe the kind of thing that in game robots like the zombie or the security guard in Quake or something might do something like this. So planning basically just means it's a little bit more than sort of shortest path planning. But it's basically like find a sequence of actions in a simple world like moving around a 2D environment that maximizes your long run expected reward. So there's a kind of utility theory, or what Laura Schulz calls a naive utility calculus, here. A calculation of costs and benefits where in a sense you get a big reward, a good positive utility for getting to your goal and a small cost for each action you take. And under that view, then in some sense-- and some actions might be costly than others, something that Tomer is looking at in infants and something that Julian Jara-Ettinger has looked at in older kids, this understanding of that. But this sort of basic cost-benefit trade off that is going on whenever you move around an environment and decide, well, is it worthwhile to go all the way over there, or, well, I know I like the coffee up at Pie in the Sky better than the coffee in the dining hall here at Swope. But to think about, am I going to be to my lecture? Am I going to be late to Nancy's lecture? Those are different costs-- both costs. It's that kind of calculation. So here let me get more concrete. So here's an example of an experiment that Chris did a few years ago where, again, it's like what you saw what the Heider and Simmel, the squares and the triangles and circles or the south gate and chibra, the red and blue balls chasing each other. Very simple stuff. Here you see an agent. It's like an overhead view of a room, 2D environment from the top. The agents moving along some path. There are three possible goals, A, B, or C. And then there's maybe some obstacles or constraints like a wall like like you saw in those movies. Maybe the wall has a hole that he can pass through. Maybe it doesn't. And across different trials of the experiment, just like in the physics stuff we vary all the block configurations and so on, here we vary where the goals are. We vary whether the wall has a hole or not. We vary the agent's path. On different trials, we also stop it at different points. Because we're trying to see as you watch this agent move around, action unfolds over time. How do your guesses about his goal change over time? And what you see-- so these are just examples of a few of the scenes. And here what you see are examples of the data. Again, the y-axis is the average human judgment. Red, blue, and green is color coded to the goal. They're just asked, how likely do you think each of those three things is his goal? And then here the x-axis is time. So these are time steps that we ask at different points along the trajectory. And what you can see is that people are making various systematic kinds of judgments. Sometimes they're not sure whether his goal is A or B, but they know it's not C. And then after a little while or some key stat happens, and now they're quite sure it's A and not B. Or they could change their mind. Here people were pretty sure it was either green or red but not blue. And then there comes a point where it's surely not green, but it might be blue or red. Oh no, then it's red. Here they were pretty sure it was green. Then no, definitely not green. And now, I think it's red. It was probably never blue. OK. And the really striking thing to us is how closely, you can match those judgments with this very simple probabilistic planning program run in reverse. So we take, again, this simple planning program that just says basically just kind of get as efficiently as possible to your goal. I don't know what your goal is though. I observe your actions that result from an efficient plan, and I want to work backwards to say, what do I think your goal is, your desire, the rewarding state? And just doing that just basically perfectly predicts people's data. I mean, of all the mathematical models of behavior I've ever had a hand in building, this one works the best. It's really quite striking. To me it was striking because I came in thinking this would be a very high-level, weird, flaky, hard-to-model thing. Here's just one more example of one of these things, which actually puts beliefs in there, not just desires. So it's a key part of intuitive psychology that we do joint inference over beliefs and desires. In this one here, we assume that you, the subject, the agent who's moving around, all of us have shared full knowledge of the world. So we know where the objects are. We know where the holes are. There's none of this false belief, like you think something is there when it isn't. Now, here's some later work that Chris did, what we call the food truck studies, where here we add in some uncertainty about beliefs in addition to desires. And it's easiest just to explain with this one example up there in the upper left. So here, and this, like a lot of university campuses, lunch is best found at food trucks, which can park in different spots around campus. Here the two yellow squares show the two parking spots on this part of campus. And there are several different trucks that can come and park in different places on different days. There's a Korean truck. That's k. There's a Lebanese truck. That's l. There's also other trucks like a Mexican truck. But there's only two spots. So if the Korean won parks there and the Lebanese one parks there, the Mexican has to go somewhere else or can't come there today. And on some days the trucks park in different places. Or a spot could also be unoccupied. The trucks could be elsewhere. So look at what happens on this day. Our friendly grad student, Harold, comes out from his office here. And importantly, the way we model interesting notions of evolving belief is that now we've got that perception and inference arrow there. So Harold forms his belief about what's where based on what he can see. And it's just the simplest perception model, just line-of-sight access. We assume he can kind of see anything that's unobstructed in his line of sight. So that means that when he comes out here, he can see that there is the Korean truck here. But you can't see-- this is a wall or a building. He can't see what's on the other side of that. OK, so what does he do? Well, he walks down here. He goes past the Korean truck, goes around the other side of the building. Now at this point, his line of sight gives him-- he can see that there is a Lebanese truck there. He turns around, and he goes back to the Korean truck. So the question for you is, what is his favorite truck? Is it Korean, Lebanese, or Mexican? AUDIENCE: Mexican. PROFESSOR: Mexican, yeah, it doesn't sound very hard to figure that out. But it's quite interesting because the Mexican one isn't even in the scene. The most basic kind of goal recognition-- and this, again, cuts right to the heart of the difference between recognition and explanation. There's been a lot of progress in machine vision systems for action understanding, action recognition, and so on. And they do things like, for example, they take video. And the best cue that somebody wants something is if they reach for it or move towards it. And that's certainly what was going on here. In all of these scenes, your best inference about what the guy's goal is is which thing is he moving towards. And it's just subtle to parse out the relative degrees of confidence when there's a complex environment with constraints. But in every case, by the end it's clear he's going for one thing, and the thing he is moving towards is the thing he wants. But here you have no trouble realizing that his goal is something that isn't even present in the scene. Yet he's still moving towards it. In a sense, he's moving towards his mental representation of it. He's moving towards the Mexican truck in his mind's model. And that's him explaining the data he sees. For some reason, he must have had maybe a prior belief that the Mexican truck would be there. So he formed a plan to go there. And in fact, we can ask people not only which truck does he like-- it's his Mexican truck. That's what people say, and here is the model. But we also asked them a belief inference. We say, prior to setting out, what did Harold think was on the other side? What was parked in the other spot that he couldn't see? Did he think it was Lebanese, Mexican, or neither? And we ask a degree of belief. So you could say he had no idea. But interestingly, people say he probably thought it was Mexican. Because how else could you explain what he's doing? So I mean, if I had to point to just one example of cognition as explanation, it's this. The only sensible way, and it's a very intuitive and compelling way to explain why did he go the way he did and then turn around just when he did and wind up just where he did, is this set of six instances basically. That his favorite is Mexican, his second favorite is Korean-- that's also important-- his least favorite is Lebanese. And he thought that Mexican was there, which is why it was worthwhile to go and check. At least, he thought it was likely. He wasn't sure, right? Notice it's not very high. But it it's more likely than the other possibilities. Because, of course, if he was quite sure it was Lebanese, well, he wouldn't have bothered to go around there. And in fact, you do see that. So you have ones-- I guess I don't have them here. But there are scenes where he just goes straight here. And then that's consistent with him thinking possibly it was Lebanese. And if he thought nothing was there, well, again, he wouldn't have gone to check. And again, this model is extremely quantitatively predictive of people's judgments about both desires and beliefs. You can read in some of Battaglia's papers ways in which you take the very same physics engine and use it for all these different tasks, including sort of slightly weird ones like these tasks. If you bump the table, are you more likely to knock off red blocks or yellow blocks? Not a task you ever got any end-to-end training on, right? But an example of the compositionality of your model and your task. Somebody asked me this during lunch, and I think it is a key point to make about compositionality. One of the key ways in which compositionality works in this view of the mind, as opposed to the pattern recognition view or the way, let's say, like a DeepQ network works-- AUDIENCE: You mean the [INAUDIBLE].. PROFESSOR: Just ways of getting a very flexible repertoire of inferences from composing pieces without having to train specifically for it. It's that if you have a physics engine, you can simulate the physical world. You can answer questions that you've never gotten any training at all to solve. So in this experiment here, we ask people, if you bump the table hard enough to knock some of the blocks onto the floor, is it more likely to be red or yellow blocks? Unlike questions of will this tower fall over, which we've made a lot of judgments of that sort. You've never made that kind of judgment before. It's a slightly weird one. But you have no trouble making it. And for many different configurations of blocks, you make various grade adjustments, and the model captures it perfectly with no extra stuff put in. You just you just take the same model, and you ask it a different question. So if our dream is to build AI systems that can answer questions, for example, which a lot of people's dream is, I think there's really no compelling alternative to something like this. That you build a model that you can ask all the questions of that you'd want to ask. And in this limited domain, again, it's just our Atari. In this limited domain of reasoning about the physics of blocks, it's really pretty cool what this physics engine is able to do with many kinds of questions. It can reason about things with different masses. It can make guesses about the masses. You can make fun of the objects bigger or smaller. You can attach constraints like fences to the table. And the same model, without any fundamental change, can answer all these questions. So it doesn't have to be retrained. Because there's basically no training. It's just reasoning. If we want to understand how learning works, we first have to understand what's learned. I think right now, we're only at the point where we're starting to really have a sense of what are these mental models of the physical world and intentional action-- these probabilistic programs that even young children are using to reason about the world. And then it's a separate question how those are built up through some combination of scientific discovery sorts of processes and evolution. So here's the story, and I've told most of what I want to tell you. But the rest you'll get to hear-- some of it you'll get to hear next week from both our developmental colleagues and from me and Tomer. More on the computational side. But actually the most interesting part we just don't know yet. So we hope you will actually write that next chapter of this story. But here's the outlines of where we currently see things. We think that we have a good target for what is really the core of human intelligence, what makes us so smart in terms of these ideas of both what we start with, this common sense core physics and psychology, and how those things grow. What are the learning mechanisms that I've just justified. Again, more next week on the sort of science-like mechanisms of hypothesis formation, experiment testing, play, exploration that you can use to build these intuitive theories, much like scientists build their scientific theories. And that we're starting on the engineering side to have tools to capture this, both to capture the knowledge and how it might grow through the use of probabilistic programs and things that sometimes go by the name of program induction or program synthesis. Or if you like hierarchical Bayes on programs that generate other programs where the search for a good program is like the inference of a program that best explains the data as generated from a prior that's a higher level program. If you go to the tutorial from Tomer you'll actually see building blocks. You can write Church programs that will do something like that, and we will see more of that next time. But the key is that we have a language now which keeps building the different ingredients that we think we need. On the one hand, we've gone from thinking that we need something like probabilistic generative models, which many people will agree with, to recognizing that not only do they have to be generative, they have to be causal and compositional. And they have to have this fine-grained compositional structure needed to capture the real stuff of the world. Not graphs, but something more like equations that capture graphics or physics or planning. Of course, that's not all. I mean, as I tried to gesture at, we need also ways to make these things work very, very quickly. There might be a place in this picture for something like neural networks or some kind of alternative pro-and-con approach based on pattern recognition. But these are just a number of the ways which I think we need to think about going forward. We need to take the idea of both the brain as a pattern ignition engine seriously and the idea of the brain as a modeling or explanation engine seriously. We're excited because we now have tools to model modeling engines and maybe to model how pattern recognition engines and modeling engines might interact. But really, again, the great challenges here are really very much in our future. Not the unforeseeable future, but the foreseeable one. So help us work on it. Thanks.
MIT_RES9003_Brains_Minds_and_Machines_Summer_Course_Summer_2015	Lecture_63_Rebecca_Saxe_MVPA_Window_on_the_Mind_via_fMRI_Part_1.txt	The following content is provided under a Creative Commons license. Your support will help MIT OpenCourseWare continue to offer high quality educational resources for free. To make a donation or view additional materials from hundreds of MIT courses, visit MIT OpenCourseWare at ocw.mit.edu. REBECCA SAXE: I was supposed to go before Ken, and thank goodness Ken insisted he went before me, because in some ways that was the most amazing introduction to my research program that you could possibly have ever had. And it articulated deeply why social intelligence should pervade our thinking about the mind and brain and the range of phenomena that people mean in social intelligence-- from extremely complex phenomena that govern the interactions of large groups of people, like war, to incredibly minute phenomena, like whether you can get your hand to a target in 100 milliseconds or less. I think that when people talk about social cognition they do actually mean all of those things. And that is both thrilling-- when you work in social cognition-- and also terrifying, especially when people are hoping for a coherent theory of all of that. I think that trying to get a coherent account of everything from your hand motions and your perception of other people's hand motions all the way to politics and sociology is daunting and, frankly, deeply unlikely. And so, by contrast to Ken-- who starts with, let's look at social interactions and see what's there, which I think is a very awesome approach-- I'm going to take almost the opposite approach, which is say, there's one thing that's probably there a lot. Let's try to study that one thing in many different ways and contexts. And the one thing, as Lou said that I'm going to talk about-- although, contrary to many people's impressions, is not the only thing I work on-- is this ability that we have. OK, so a little demo of the problem that I work on-- and because it's early in the morning and everybody needs to wake up, I'm going to get you guys to do this as a task, so as an experiment. So in this experiment, I'm going to ask you guys to make a moral judgment of a character. Her name is Grace. And the way you're going to make a moral judgment is, I'm going to tell you something she did, and you're going to say how much blame she deserves-- moral blame, how wrong that was. You're going to do so by raising your hand. The more wrong it was, the higher your hand goes. And everybody has to vote. OK? Yes? OK, so this is a story about Grace. She's on a tour of a chemical factory. So they're walking around being given a tour. There's a break in the tour. And she goes to make coffee. Another girl on the tour asks for a cup of coffee with sugar in it. So Grace goes to the coffee machine to make a cup of coffee for herself and for this other girl. Next to the coffee machine is a jar of white powder labeled sugar, so Grace thinks the powder is sugar. She puts some of that powder in the other girl's coffee. But it turns out that powder is contaminated by a dangerous toxic poison, and when the girl drinks the coffee, she dies. How much blame does Grace deserve for putting the powder in the coffee? OK, now what if I slightly changed that story? So next to the coffee machine there's a jar of white powder and it's labeled dangerous toxic poison. So grace thinks that the powder is toxic poison. And she puts some of the poison in the coffee, and when the girl drinks it, she dies. Now how much blame does Grace deserve for putting the powder in the coffee? So what's characteristic about these stories is that, in the story I told you, everything was the same from the beginning, the scenario where Grace was, to the action and the outcome-- that the girl died. But your moral judgments differed by about the entire scale that I gave you, from saying that she deserved almost no blame to saying that she deserved pretty much as much blame as you could reach. And that's the same kind of moral judgment we get from typical human subjects and also from MIT undergraduates, which say that in scenarios like the one I gave you, what matters most for the moral blame that we assign is not what happened-- did somebody die or not-- or how bad that outcome was. But it's what Grace thought she was doing, whether she thought the powder was sugar or she thought that it was poison. I should just say right away that I set up that scenario. I gave you the best case scenario for the role of beliefs. It's easy to make these things way more complicated. But that scenario isolates one important feature of our moral judgment and also an important feature of a lot of the rest of our social cognition. It's not how we avoid bumping into people in subways, but a lot of the other kind of social cognition we do about the people that are around us, which is our ability to assign thoughts or internal mental states to other people. So in psychology, this ability has been studied from kind of relatively simple perceptual phenomena like assigning intentions and goals to simple moving characters in an animation. This is the very famous Heider and Simmel example from the '40s. This ability has been studied all the way to understanding some of the most complex, abstract ideas that we ever encounter, like the famous apocryphal statement attributed to Alan Greenspan, which is, "I know you think you understand what you thought I said, but I don't think you realize that what you heard was not what I meant. " So to the degree that our minds let us make any sense of that at all, we're using our ability to make sense of other people's minds. How many people here have seen the standard test of this ability of thinking about other people's thoughts, which the false belief task? How many people have seen somebody do a false belief task or give a false belief task? How many people would like to see a false belief task? OK, so then I'm just going to show you one. So as I said, the scope of tests of our ability to think about other people's thoughts or internal states is very large. I'm saying that two ways on purpose because actually, although these are often conflated, I think there's a really important difference between thinking about epistemic states-- so things like what you know, what you see, and what you think-- versus states like what you want and how you feel. I think it turns out empirically that those are really different problems. And I'm almost exclusively going to talk about the first one, so how we think about what other people see, think, and know-- but not want or feel. At the end I'll come back to wanting and feeling. OK, so how do we know what other people have seen and what they think and what they know? This problem was set up as kind of a litmus test for our ability to think about other people's minds, starting in the late '70s and coming out of comparative psychology. So the origin of this problem for psychology is, everybody knows humans could do this. What about animals? And actually, the debate about whether this capacity for thinking about other people's thoughts is or is not shared with which other animals has gone on continuously since the late '70s and has not been resolved. That's the origin of this debate, and it's not resolved yet. But it led to the construction of this particular task as a litmus test for what one person knows about somebody else's thoughts, called the false belief task. And so here's what a false belief task looks like. This is being given to a five-year-old human child. This is the first pirate. His name is Ivan. Do you know what pirates really like? CHILD: What? REBECCA SAXE: Pirates really like cheese sandwiches. CHILD: Cheese? I love cheese! REBECCA SAXE: So Ivan has his cheese sandwich, and he says, yum, yum, yum, yum, yum. I really love cheese sandwiches. And Ivan puts his sandwich over here on top of the pirate chest. And Ivan says, you know what, I need a drink with my lunch. So Ivan goes to get a drink. And while Ivan is away, the wind comes, and it blows the sandwich down onto the grass. And now, here comes the other pirate. This pirate is called Joshua. And Joshua also really loves cheese sandwiches. So Joshua has a cheese sandwich, and he says, yum, yum, yum, yum, yum-- I love cheese sandwiches. And he puts his cheese sandwich over here on top of the pirate chest. CHILD: So that one is his. REBECCA SAXE: That one's Joshua's. CHILD: And then his is on the ground. REBECCA SAXE: Yeah. That's exactly right. CHILD: So he won't know which one is his. REBECCA SAXE: Oh-- so now Joshua goes off to get a drink. Ivan comes back. And he says, I want my cheese sandwich. So which one do you think Ivan's going to take? CHILD: I think he's gonna take that one. REBECCA SAXE: Yeah, you think he's gonna take that one. All right, let's see. CHILD: I told you. REBECCA SAXE: Oh yeah, you were right. He took that one. OK, so that's called passing the false belief task. And the thing that's reported in scientific papers is that he correctly predicted that Ivan would take Joshua's sandwich. Although if you watch the video, you see that the knowledge the kid is bringing to bear is a way richer than just his correct prediction and includes him, in fact, trying to stop me in the story to warn me of what's coming. So it's a rich interpretation of what other people know and don't know and will know and haven't seen and so forth. The reason why this task became so famous is that not all participants perform the same way. And so one class of participants who've become the focus of intense scrutiny is slightly younger kids, namely three-year-olds. So I'll give you a sense of what that looks like. This is a three-year-old. He's paid equally rapt attention throughout the entire story. And we come to the crucial moment, and he's asked again the same question. And Ivan says, I want my cheese sandwich. Which sandwich is he going to take? Do you think he's going to take that one? Let's see what happens. Let's see what he does. Here comes Ivan. He says, I want my cheese sandwich. And he takes this one. Uh oh-- why did he take that one? OK, and so the traditional read of what just happened there is that's a kid who gets wanting, right. Ivan wants his cheese sandwich. But he doesn't get believing. He doesn't understand that because Ivan left his cheese sandwich on top of the pirate chest and he doesn't know that it's been moved that he'll believe that that sandwich is his, and that his actions depend on his own beliefs-- his internal representation of the world, rather than the true state of the world, namely which one is his cheese sandwich. And that's the source of both this wrong prediction-- why does he say that he'll take his cheese sandwich-- and the wrong explanation. So when he goes to take the other cheese sandwich, the one that's actually Joshua's, then we say, why did he do that. And again, this is typical performance that the little kids confabulate. They come up with a reason why he might have taken that other cheese sandwich which is consistent with him not wanting his own anymore. So in this case, it's that his fell on the ground. He doesn't want his anymore. That's why he's taking Joshua's sandwich. And that pattern of performance was interpreted as evidence of conceptual change and development-- kids going from having a partial understanding of other people's minds that included wanting to a richer interpretation of other people's minds that also included believing. So what I want to get from this actually is not whether or not it's true that there's conceptual change between three and five, although I do think it is true, but just an idea of what capacity are we talking about. We're talking about the capacity actually that the five-year-old showed, however they got it and whenever they got it. It's this capacity to, when watching other people act in the world, bring to bear-- both spontaneously and when asked-- a conception of the other person having beliefs, perceptual history, knowledge, an internal representation of the world that guides actions. And so that is what I'm going to call thinking about thought. And the idea that this is a domain that you could study on its own-- well, there's two questions here. One is can you study this at all. And the second one is can you study it separate from the whole rest of cognition. Both of those are related to Liz, and indeed Nancy, and many people's worries that you could never make progress on a problem like this, which I share. I share the worry that you could never make progress on this. And so what I want to tell you guys is two phases of my attempt to make progress on understanding how we do that. How do we think about other people as containing internal mental lives, mental representations. I'm going to talk about just fMRI, although I do use other methods to study this problem. But I think fMRI has been both an incredible gift to our ability to understand the human mind and also imposes a huge number of limitations on what we can discover. And so what I'm going to tell you about is just a tiny bit of my phase one investigations using the early strategies that fMRI allowed us and then a more in-depth look at how I'm using more modern techniques in fMRI to try to get further. This is partly because I think it's interesting what we've learned. But it's mainly because I think that you guys might not actually want to know about theory of mind, but you might want to know if you can fMRI to study interesting questions about the human mind and how. And so I'm going to focus on three ways to use modern techniques in fMRI to study interesting representations in the human mind, hoping that either you'll learn something about theory of mind or something about how you could use fMRI to pursue your own interests. So phase one in fMRI, which as Liz said started 15 years ago, is-- OK, thinking about other people's thoughts, is that a thing in the mind and brain at all? So when you go to start studying something, you want to know, am I studying a part of a problem, or am I just studying the whole mind, our entire capacity to think any interesting, complicated thought. And fMRI turns out to be more useful, mostly, when you're studying something that is in some way compartmentalized from the rest of the mind. And so one question is-- is theory of mind, the ability to think about other people's thoughts, in any sense its own problem? Or are we just studying the whole problem of human intelligence and capacity? So that was sort of the first question that we set out to answer. We and a number of people did this. And the way we did it is that we had people in an fMRI machine doing basically an adult version of the pirates task that I just showed you. So they read short verbal stories that describe somebody who comes to have a false belief. This is an example. So Ann puts lasagna in the blue dish. Ian takes the lasagna out and puts spaghetti in the blue dish. And then we ask, what does Ann think is in the blue dish. OK, so this is a very simple encapsulation of our ability to represent what somebody else thinks and separate it from the state of the world. So while you were doing that, you were clearly using your theory of mind. But you were clearly also using many, many, many other capacities of your mind and brain, like the capacity to see those words, to know they are words in English, to put them together in sentences, and then to make a response by pushing a button. So we're using everything from your eyes to your fingers and most of the brain in between. And then the question is, the part that required you thinking about thoughts-- is there any sense in which that's special or different from the whole rest of the logical and cognitive capacity of your brain? So to ask that question we designed a control condition in which you similarly read stories that involve something that was true and becomes false. You need to think about those two and respond using a button press. But in this case, what it is is a state of the world. So this is an island and a photograph taken of it. Then the photograph, of course, stays the same. But the world changes. So there's a volcano that erupts. And now we can ask you either about the photograph, what's in the photograph, or what's the world actually like now. And the idea is that in this comparison you need the ability to see the stimuli, read English, put together your logical thoughts, and choose a button press in both cases. But only in the first case do you also need to think about other people's thoughts. And so that comparison would let us look for brain regions where blood oxygenation or metabolism is higher if you're thinking about other people's thoughts. So that is old news now. The simple answer is that we and many, many other groups that tried this in many different ways found a whole group of brain regions where metabolism or blood oxygenation is higher if you need to think about other people's thoughts in the stories. Part of what's interesting though about this brain region is not just the claims about selectivity. The other thing that's interesting-- and extremely fortunate for research purposes-- is that the signal is ridiculously strong and reliable. The difference between thinking about somebody else's thoughts and other logical problems-- in terms of how significant, how reliable, how similar across individual subjects-- is comparable to the difference between looking at gradings and not looking at gratings in V1, which is nuts. That's crazy that something this complicated and abstract would have an unbelievably large, robust, reliable signal in individual subjects. I'll give you a little hint of it. But everyone who has ever come through my lab says that they never believe me until they see it in their own data. And you can do this in any individual subject. So here's just three individual participants after five minutes to 10 minutes of scanning. You need to read only between 10 and 20 total stories in literally five to 10 minutes. And every individual subject basically shows the same pattern of brain activation for thinking about thoughts compared to the other stories. It's just this unbelievably strong signal, literally unbelievable. It should not possibly be true on any a priori story, except for maybe the story Ken just told you about how social cognition is the fundamental basis of everything. When you look inside this brain region, this is in one of them. I'm showing you pictures of the right TPJ. It's one of five cortical regions. I'm going to talk a lot about it, because the data from the right TPJ are particularly clean. So in the right TPJ, that's average percent singal change in some of our early experiments to stories about beliefs compared to control stories about photographs. Two things are striking. One is that it's a really big difference-- a big positive signal when you're reading stories about beliefs, and not much when you're reading stories about photographs. The other thing is that it starts at the time you start reading the story. So you start reading a story, and the signal starts to go up. This is just showing that difference in how much you think about thoughts contributes a lot of variance across many different individual stories. And if you look within the story, it's the time when you're thinking about a thought that you get activity in this brain region. We also spend a bunch of time saying fMRI, as everybody knows, is a correlational signal. Does this brain region actually play a causal role in letting you think about thoughts? And so we did a version of the same experiment that I gave you guys on moral reasoning with TMS and asked whether using TMS on the right TPJ compared to a control brain region would disproportionately affect how you use people's mental states in making moral judgments. We showed that after TMS to the right TPJ compared to a control brain region, people use the beliefs of the character less in making their moral judgments. And so where we get to after all of this is a hypothesis. This was after about eight years that I was saying here's what we've learned. We've learned that the right TPJ is selectively involved in theory of mind, and so selectively depends on all the experiments I didn't show you. That's a claim about specificity. But "involved in"-- that's a euphemism. And it's a euphemism that I think a lot of cognitive neuroscientists use and are satisfied with. But after a while, I found it deeply embarrassing, like-- what on earth is "involved in"? And so what I want to talk to you about is how to get beyond the euphemism of "involved in" in using fMRI to understand the mind. This is what fMRI in my hands typically looks like. You read a bunch of stories in the scanner, and we record activity in the brain region-- here, for example, the right TPJ-- while you're reading those stories. And so our traditional measures are the measures that let us estimate specificity and selectivity and answer all the questions you guys asked me. Is it more for this, less for that? What makes it go up? What makes it go down? Those measures, called univariate measures, measure the amount of activity in that region, on average, as you read a different story. So you get something that looks like this. And what you show is that this brain region responds a certain amount. Or there's a certain amount of activity metabolism in this brain region while you're reading that story. And so what we do with that is we make arguments about selectivity and these kinds of things that we've been talking about this entire time. And if you did that in the reverse direction, you'd say, OK, what can we learn about these stimuli or the representation of these stimlui-- these two stimuli-- from activity like this? Well OK, both of them are within the set that this brain region cares about. They both elicit high activity. So both stories involve thinking about thoughts. And one of the things we showed early on is that activity generalizes in the sense that many different stories about many different kinds of thoughts all illicit activity in this brain region. And so from the amount of activity in this brain region, you know something like that story is about thoughts. And I told you that I think that that is related to this idea of involvement. This brain region is involved when a story describes thoughts. OK, what's wrong with that for making theoretical progress on theory of mind is that, with respect to the representation of other people's thoughts, that doesn't tell us anything about how our brain does it. So for example, what it doesn't tell us-- it doesn't tell us how we know who thinks what. It doesn't tell us why they think that. It doesn't tell us what the consequences were of them thinking that. It doesn't tell us how our brains track or represent any of these properties. So the things that would make something a theory of mind-- a representation of who thinks what, why, and with what consequences-- we can't see in the univariate signal. So what I would like to make progress on, what I think we're starting to make progress on using MVPA, is getting beyond that this brain region is involved in theory of mind and trying to ask something about what is represented in this brain region. And we're doing this using a key assumption which comes from systems neuroscience, which is that we can think of representations in terms of population codes of features or dimensions. And I want to say that right now because that is an old, discredited theory of concepts, but nevertheless a powerful strategy in neuroscience, including in this context. It's another thing I could talk at greater length about. So the idea that we're going to look at is that populations of neurons will respond differentially to features or dimensions of our stimuli. And by figuring out what the main features or dimensions are of our stimuli, we can infer something about the representation underlying-- the representation that this brain region participates in-- and that is the representation of theory of mind. OK, So what is MVPA? I'll briefly say my idea of how to think about MVPA. So a traditional analysis-- the things that we were doing mostly for the first 15 years of fMRI-- are called now univariate analyses. You would take a patch of cortex, as represented by a bunch of pixels in the brain-- they're called voxels, a bunch of volume elements in the brain you're studying-- and look at the average amount of response. The unit of analysis was the amount of response. These experiments typically proceed in what's called now the forward or encoding direction. So that is, you have some hypothesis of what might be represented. You vary it in your stimuli. And you look at how varying that dimension in your stimuli causes differences in the magnitude of the thing that you're measuring. What was most effectively revealed by these analyzes are differences in the cortex at the scale of regions, what one region as opposed to another region does, so what the kind of large-scale structure of the cortex is on the scale maybe of a centimeter. And that turns out in many contexts-- especially in the back half of the brain, the representation regions-- to correspond in some sense to the stimulus type. What kind of thing were you dealing with? What is it that you're looking at or processing? And then this is, I think, in some ways the shortest possible version of Nancy's amazing 30-year research program of figuring out how to parcellate cortex into chunks of about a centimeter that correspond to something about the type of stimulus that we're presenting to you. And divide up in this forward direction. Think of a type of stimulus. Find the brain region where the magnitude of response is selective to that stimulus type. MVPA analyses-- so multivoxel pattern analysis-- are contrasted to this in the sense that they tend to be multivariant. So that is, you're looking at not how much on average a group of voxels respond, but the relative response between one voxel and another from trial to trial. So you're looking at which of two voxels is higher or lower than the other, rather than what their overall amount of activity is. It has mostly, though not always, been used in the reverse or decoding direction. So the answer at the end is-- given that I got this pattern, what can I figure out about the stimulus? So that's the way many of these analyzes proceed. You ask, having done all of this, now I get a new pattern of activity. What can I decode about the stimulus from the new pattern of neural activity? To me, these analyzes are most interesting when they're looking for things smaller than a region. This is again another interesting long conversation that I would have got to at the end. All the mathematical techniques of MVPA could be used to rediscover all of the things Nancy already discovered using the traditional analyses. And in fact, if you use them uncarefully that's what you're most likely to do, because those are huge signals in the brain. And so if you're not careful, what you will do is just re-go over old territory with new math. I am more interested in these techniques when they let us see things we could never see before. So when, instead of telling us about region level differences or centimeter scale differences, they're telling us about much smaller and more interleaved populations on the spatial and representational skills and when what they're revealing are not the type classifications of stimuli-- so the things that decide whether this region or that region will be more activated-- but for a given type of stimuli, what are the key dimensions of representation. So the reason why I think MVPA is giving a new life to fMRI is because many of the most interesting questions about cognition and cognitive science that we wanted to answer and that fMRI never let us answer were about within-stimulus type features or dimensions. What makes this face look like person A versus person B? What makes this thought predict moral blame versus not blame? So within-type dimensions of importance-- and MVP I think is letting us do that in its most interesting applications. The intuition here is that-- think about a region, like the right TPJ, or the face area if you think about faces. Or V1 is often where I start, because we know enough about V1 that I can use it to imagine what we're talking about. So you can think about that whole area. And you think, what can we learn about what it does. So let's talk about V1. Does everybody here have some sense of V1? Everyone's had kind of a first introductory neuroscience class, OK. So V1 is called V1 because information goes from your eyes to the LGN of your thalamus to V1. It's the first cortical stop of visual information. And one way that we know that it's very involved in vision is that if you were doing visions, if you're seeing visual stimuli, you get a big response in V1. If you're not seeing visual stimuli, like you're hearing auditory stimuli or feeling tactile stimuli, you don't get a big response in V1. So that's a selectivity type measure. It's a univariate measure for the amount of activity in VI that tells you V1 is in some way involved in vision, relative to audition or somatic sensation. But that misses pretty much all the interesting contributions that visual cortex makes to vision. What we want to know about V1 is not that it is involved when you are doing vision and not involved when you are not doing vision. We want to know what transformations over the information coming from LGN is V1 implementing-- what computational transformations, what representations. And that's why theories like Marr's theory-- which say that it's edge detection, or that there are receptive fields, that it depends on the contrast, the position, and the orientation of the information in the field that counts as an account of the representation that V1 forms of the image in the first bottom-up sweep. In a way, that's saying "it's involved in vision" doesn't even begin to count. OK, so the question is, if we were going to look at V1, could we discover from fMRI that V1, for example, has an orientation map, that neurons in V1 have an orientation preference? That's a key feature of neurons in V1. It's a key feature of the computation V1 does. Different from the LGN and the retina is the orientation map, a preference for the orientation of a contrast and edge. And the answer in standard analyses is-- no, you can't, because V1 as a whole will activate to big images regardless of the orientation of the content of the image. So you need to be able to get to something more fine-grained than V1. You need to be able to say there are different subpopulations of neurons inside V1, some of which will be responding when a line is like this, and some of which will be responding when a line is like that. And that's the decoding perspective that says, if we wanted to look at V1 and know is the line like this or like that, the way we would tell is not how much activity there is in V1. But is there relatively more activity in the population of neurons that responds like this, or in the subpopulation of neurons that responds like that? And it's the relative activity in those two populations that would let you say, is the line like this, or is it like that. That's population coding or population decoding. And then you take that to the fMRI level. So now you want to say, can we tell which of those two subpopulations is more active in fMRI? Now, if you could measure the individual neurons-- so if you know these neurons prefer this and these neurons prefer that, and then I measure your firing patterns-- then decoding from the population is simple. What makes it really hard in fMRI is that the unit of measurement is the blood oxygenation in 100,000 neurons, 200,000 neurons, maybe 500,000 neurons. And so it seems potentially really unlikely that you would be able to tell from the fMRI signal whether the neurons that prefer bars like this or bars like this are more active, because they're all intermixed inside a single measurement in fMRI. And so it's not stupid that we used to focus on things like how much activity. The reason we used to focus on how much activity with fMRI is that it was quite plausible that that's all fMRI could tell us. The neural populations, like orientation preferring neurons in V1, were too spatially mixed to tell the difference between them in fMRI. And so what we were going to get was just how much activity in the population as a whole. So the traditional way of thinking about what you got out of fMRI is, yes, you would see differences across voxels, so these fine spatial patterns. But there's so many things that could cause fine spatial patterns that we don't care about-- noise, for a start. Where the blood vessels happen to be is another thing. And so people assumed, I think very reasonably, that because fMRI is such a core spatial measure, that the only thing it could tell you was the average over the millions of neurons that make up a region. And there's a key intuition underlying MVPA. So there's the big signal which is the regional signal-- V1 and vision-- and there's lots of noise. But there might also be inside there a tiny bit of spatial pattern that says something like-- well, this voxel happens to have more neurons that prefer one orientation. And this voxel happens to have more neurons that prefer a different orientation. And so from the relative activity in those two voxels, we could still tell you the orientation-- even though that would be a tiny, subtle little signal superimposed on top of this massive signal, which is the average of V1. That was the intuition behind multivoxel pattern analysis when it was first proposed. And it's now sweeping the fMRI world, many different versions of these analyses. And so actually what I'm going to do again-- to give you a more concrete sense of what this is and how it works-- is I'm just going to show you two different ways MVPA is done concretely in my lab to try to get you more of a sense of what's going on. And we can come back to these more general issues of what it's measuring and what that means. OK, so here's what it looks like when we do MVPA. Again, if it helps, think about the analogy from vision. We've gone from saying, is this vision or audition, to trying to say which orientation is it. So we're moving from saying, is this theory of mind or not, to trying to say anything interesting about the space within theory of mind-- some dimension that might matter within theory of mind. And the first dimension or potential feature that we wanted to look for we chose because it really matters to human judgments. And it's the one that you guys did in the very beginning of my talk-- telling the difference between somebody who knowingly or unknowingly commits murder. That, as you saw, makes a huge, huge difference in behavior. And also, we know it's represented in the right TPJ because of the TMS experiment. If we mess up the signaling in the TPJ, we change that judgment. And so that was our best guess, that if any feature of other people's mental states is represented in the right TPJ, it would be that feature. If MVPA was ever going to be able to decode a feature of other people's mental states, we should start there. That was the idea. OK, so here's how these experiments go. In every every trial you read a long, complicated story that sets up a murder. So here's an example. Your family's over for dinner. You want to show off your culinary skills for one of the dishes. Adding peanuts will bring out the flavor. So you grind up peanuts and put them in the dish and feed everyone. Your cousin, one of the dinner guests, is severely allergic to peanuts. You had absolutely no idea about his allergy when you added the peanuts. And then at the end we ask how much blame you should get. Somebody asked me this earlier. This is in the second person and doesn't matter. Somebody asked me if you could do it in the second person, and you can. What's nice about this experiment is that we can do a relatively minimal pair. So in all of our old experiments we wrote one set of stories about people's mental states and a completely different set of stories about other things. And those stories are different in many, many, many ways. In this experiment, we make one tiny change. So we make, for example, a change from you had no idea to you knew. We change on average two to four words in this whole long scenario. So we can make these tiny interventions. It's a complicated stimulus, but the change we make is very small and totally changed the meaning of the whole story by just changing your mental state. OK, what univariate analyses would say is, this is a really important fact. Whether you knew or you didn't know about your cousin's peanut allergy is really important to the moral judgment of what happened. We know that. And it's represented in the right TPJ, because if we TMS the right TPJ, you make your moral judgments of this distinction specifically change. But if you ask how much does the right TPJ respond to these stories, the answer is the right TPJ responds exactly equally to these two conditions. And the intuition is, because in both cases it matters what you think. It matters that you knew, and it matters that you didn't know. And the right TPJ is tracking the important information about what you think. And so it's activated for both of these kinds of stories. So that's a univariate analysis. Now what's a multivariate analysis? So here's the key intuition behind a multivariate analysis. The idea is, think in a very abstract similarity space. If we take the two stories-- and so we take the story you had no idea about your cousin's allergy when you added the peanuts. That story is complicated. It has many important dimensions. Now we take a new story. This is a story about, for example, a faulty parachute. Within that story there's many, many different dimensions. It's about parachutes. There's all kinds of complicated things going. But there's this one feature-- whether you knew or didn't know that the parachute was faulty. There's another story about publicly shaming your classmate by saying something embarrassing about their essay. So again, that's a whole new scenario with all kinds of dimensions in it. But there's this one feature. Did you know or not know that the person who wrote the essay was in the room when you said that publicly shaming thing? A different story is about demonstrating your karate skills and knocking out your classmate-- again, totally new moral scenario. But again, this one feature-- did you know or not know that your classmate was there when you did the kick? Now here's the idea. Even though each of those new scenarios is completely different, if there are different subpopulations within your right TPJ responding when you knew you were going to cause harm-- compared to when you didn't know you were going to cause harm-- then even though the pattern of activity in your right TPJ will be different on every trial-- because you're representing a different person having a different mental state in a different context-- a little part of that response will be the same. Or it will be different in the same way, right, because the same cell population will be more active for all the stories that have knowing harm. And the other population will be relatively active in all the stories that have the unknowing harm. And so the logic is that if we could look in the right TPJ and measure the pattern of activity-- and hope that reflects something like the relative activation of different cell populations inside the right TPJ-- that the pattern of activity would be more similar for pairs or subsets of stories that share this one feature, and are different in every other way, compared to pairs that are different in every other way and don't share that feature. OK, so this is the central logic. Take any two stories within the set. They're all unique. So those two stories that are different, you're representing a new mental state of a new person. You have a new pattern in your right TPJ. But if they share the feature that you knew you were going to cause harm, that would be something a little bit similar in your right TPJ activation compared to if they don't show that feature. Does that logic make sense? And so what you get is a spatial pattern of activation. So we're now not looking at how much the right TPJ responded. But within the space of the right TPJ, where was there a little bit more or a little bit less activity? And these signals are tiny compared to the thing I showed you before. So the amount of activity in the right TPJ is a big signal. The relative activity between one voxel and another is a tiny signal. And it's superimposed on a lot of noise. But if there's anything there at all, then you'll still be able to pick up a little more similarity for pairs that are matched on the feature of interest compared to pairs that are not matched on the feature of interest. That's the logic behind a Haxby style analysis. And so literally what you do is, you take the vector of responses across all the voxels inside a region, and you correlate them across subsets of your data. And you ask whether the correlation in space-- so what it looks like, the spatial pattern of activity over those voxels-- is more similar for pairs that share the feature you're interested in compared to pairs that don't share the feature that you're interested in. And what you get at is two numbers-- the correlation for pairs that do share the feature and the correlation for pairs that don't share the feature for each individual subject. And the question you ask in a Haxby style correlation is what's called the within-condition correlation. So the spatial correlation of the response to two independent sets of stories that share this one feature-- is the spatial pattern more similar in that pair compared to a pair that don't share that feature, when everything else is different? And so what you get out of an analysis like this-- for example, in our first attempt to do this in these stimuli-- there's these two correlations. There's the within-condition correlation and the between-condition correlation, and you ask if they're different. OK, and what we got in our first experiment is that the within-condition correlation is significantly but a tiny bit stronger than the between-condition correlation. So there's a lot of things to ask about this. But the first question is-- is that real, or is that a coincidence? That is the first thing you want to know when you see data like this. Afterwards, we can ask what does it mean. But let's start with is that real. And so the way that you ask is it real is, you just make sure that it would replicate, that in independent data you'd get the same answer. And so just before we set out to actually replicate this experiment, we remember that we had actually already run this experiment two times before-- because we were studying this process of representing accidental and intentional harms for a long time before we thought of using MVPA. So we had these two old data sets in the lab, two whole independent experiments in which people had read stories about knowing and unknowing harm. And the other thing is that we had manipulated this distinction in different ways across the stimuli. So in the example I just told you, the way that we did it is, we said you knew about the allergy or you didn't know about the allergy. But in the older experiments, like this example I gave you at the beginning of the talk, we had described two different beliefs-- so either believing that it's sugar or believing that it's poison, so no negation. This is just important because that's a different way to create the same distinction. And what you want to know is, are you decoding the abstract thing-- that she knew she was causing harm or not-- or something less abstract, like whether the story has negation in it. That's an alternative possibility. And so in experiments B and C, we had done it this way. It's also in the third person, not the second person. So if we find the same result, then it generalizes across all these incidental features of the way the experiment was run. OK, that's experiment two, and that's experiment three. I also want to say that there's some weird magical property of being a scientist, where if you don't know the hypothesis when you're running the experiment and you have all the data and then you go back and check, there's something more real about it than if you knew the hypothesis before you ran the experiment-- even though that makes no sense whatsoever. There's just this experience like, if I had the hypothesis in my head, maybe it somehow got from my head to the data. But when the data were already there and then you went back and analyzed them and the effect was hiding in the data that you'd had on your server, there's something way more real and magical about that. So anyway, because it was there in all of our old data, I just believed it was true. The other thing to notice about this is, to get an MVPA signal, we didn't change anything about the fMRI. We didn't change the resolution-- the temporal resolution, the spatial resolution. You can know that for sure, because these are our old data that we had before we started doing MVPA. MVPA is not a technique for collecting better data. It's a technique for getting more information out of the same data. It's an analysis technique. It's a way of thinking about data, not a way of getting data. OK, so what this says is that however similar two unrelated stories are about a case in which somebody kills somebody, they are more similar if they are both cases of knowing murder or both cases of unknowing murder than if you cross that feature. So just making that future match makes the pattern of neural response in the right TPJ more similar, suggesting that which part of the right TPJ is more or less active contains information about whether or not the person who committed the murder knew what they were doing at the time. This is specific to the right TPJ. So these are a bunch of the other brain regions involved in theory of mind and social cognition. And none of them contain any information about this dimension at all. So this dimension is represented in the right TPJ and not represented anywhere else. There's another thing that makes these data interesting. People are reading these stories, and they're making moral judgments. And moral judgments of these stories vary across people. So some people tend to go more with what the person thought, whereas other people tend to go more with what the person caused. It's not extreme individual variability. Everybody agrees that it's worse to knowingly murder than to unknowingly murder. But there is variability in how much worse. Some people think that basically what you thought you were doing is all that matters in these stories, whereas other people think both of those things matter. So it matters to some degree that you caused the murder and to some degree that you didn't know. So there's individual variability. And one thing that we can look at is, how does the individual variability in the behavior relate to the individual variability in the representation. So what this looks like is, on the x-axis I measure-- for you, how much worse are intentional than accidental harms. How much worse is it when you knew you were going to cause harm than when you didn't know? So that's always going to be a positive number. Everybody thinks it's worse. But for some people, it's a lot worse than it is for other people. And then relate that to, while you were reading that story, how different were the patterns in your brain when you were reading about knowing harm compared to unknowing harm. Does that make sense? They're pretty correlated. So the more that you represented knowing harm as different from unknowing harm in your right TPJ, the more that you judged them as different when we asked you for moral judgment. And the pattern difference in your right TPJ accounts for 35% of the variance in your moral judgment, which is pretty amazing, because that's a pretty noisy measurement. Actually it's both. It's a pretty noisy measurement of your brain and a pretty noisy measurement of your behavior. So it's quite amazing that those are that correlated. So that's what's cool about the data. But we'll get to the method. So Haxby style correlations-- these are called Haxby style because they were the first form of MVPA introduced, and they were introduced by Jim Haxby in 2001, so actually a long time ago. It took a long time for other people to recognize what a cool technique this was. But he had this idea a very long time ago. And the idea is, take a region you care about and ask this basic question. For some future that I wonder if it's represented, is the correlation across neural responses more similar when the stimuli share that feature than when they don't share that feature? So that gives you a pretty robust measurement, because you're using all the voxels in the region to get one number out-- the correlation. And you're doing it over partitions of the data, often halves of the data, so many trials are going into both the train and test. So in this case we're using halves of the data, even halves and odd halves. And so each of the things we're correlating is a relatively less noisy neural measure because it has many trials averaged into it. So it's robust and simple. In this case, it can be sensitive to pretty minimal stimulus variations. As I showed you, this is a two- to four-word variation on an 80-word story. So it's sensitive to small distinctions in the stimuli. Here we showed that it generalizes. So we used totally independent stories in the train and test set. And so we're always generalizing from one set of examples to a totally different set of examples. It gave us a measure that was stable within a participant in the sense that the measure in each individual related to that individual's behavior. So it's characterizing individuals in a relatively stable way. And we could show that it differs across regions. So we could show that this was present in the right TPJ but not present in other regions. And that's a bunch of stuff you would want to know. That's a whole bunch of extra information than we ever were able to get before. And I'll give you one more example of the way that Haxby correlations can be used. So in this case I showed you, we hypothesized one dimension. And we tried to decode that dimension. Obviously, you don't only have to do one. And so another way to do this is to build stimulus sets that, for example, have two orthogonal dimensions and ask about both of them. SO here's an experiment in which we asked about decoding two orthogonal differences within the same set of stimuli. So again, you're reading stories about people who are having experiences. And some sets of these stories vary. So here's a bunch of stories. Leslie has just been in a big, important interview. And he sees himself in a mirror, and he sees that his shirt has a big coffee stain down the front. And another one is-- Eric gets to a restaurant to meet his fiance's parents, and he sees them and they're looking happy So that's two completely different stories. And then the third story-- Abigail is painting her dorm room, and she hears somebody's footsteps down the hallway. And the footsteps sound like her beloved boyfriend's. So these stories are all different. Again, they're all But the first two stories I read you share a feature, which is that somebody in the story is seeing something. And they don't share that feature with the third story, in which somebody in the story is hearing something. Does that makes sense? Compared to, for example-- Quentin hears a phone message, and the message says she has bad news to tell him. That's another story that shares this feature that somebody in the story is hearing something. And so we can use this set of stories to ask, is the neural response to stories in which somebody is seeing something more similar within that set? So one set of stories about seeing something is compared to another set of stories about seeing something. Are those stories more similar to one another than when you cross that feature? So you ask one set of stories about seeing something compared to one set of stories about hearing something. And so in the right TPJ, what we found is that stories about seeing are more similar to other stories about seeing. And stories about hearing are more similar to other stories about hearing than when you cross that feature. But, as you may have noticed, the stimulus set had another distinction in it, which is whether the thing is good or bad that's happening to you. So finding out after an interview that you have coffee down your shirt or hearing a message that says there's bad news, those are both bad things. Whereas seeing your fiance looking happy or hearing that your beloved boyfriend is coming down the hallway, those are both good things. And so we could ask in the same dataset, what about stories that share this feature of valence. The pairs of stories that are matched on valence, do they have a more similar neural signature than the pairs of stories that are crossed on valence? And in the right TPJ they're not. We've actually found this a whole bunch of times. The right TPJ doesn't care about valence. Other regions do-- don't worry-- we do represent valence. But the right TPJ doesn't represent valence. So that's another way that you can use this method-- hypothesize two or three orthogonal dimensions within the same stimulus set. And then we can get, for example, interactions between these to say, OK, the right TPJ does represent some dimensions, doesn't represent other dimensions-- in principle. So you can test potentially multiple orthogonal distinctions. There's a whole bunch of limitations of Haxby style correlations. One of them is that all the tests are binary. The answer you get for anything you test is that there is or is not information about that distinction. There's no continuous measure here. It's just that two things are different from one another or they are not different from one another. And so once people started thinking about this method, it became clear that this is actually just a special case of a much more general way of thinking about fMRI data. This particular method, using spatial correlations, is very stable and robust. But it's a special case of a much more general set.
MIT_RES9003_Brains_Minds_and_Machines_Summer_Course_Summer_2015	Unit_3_Debate_Tomer_Ullman_and_Laura_Schulz.txt	The following content is provided under a Creative Commons license. Your support will help MIT OpenCourseWare continue to offer high quality educational resources for free. To make a donation or view additional materials from hundreds of MIT courses, visit MIT OpenCourseWare at osw.mit.edu. TOMER ULLMAN: What we're going to do in the last section is not so much have a lecture as a debate. And it's going to be a debate between myself and Laura Schultz. And hopefully at the end, all of you can join in for some sort of free for all. Although, of course, you're welcome to ask questions at any moment. Now, Laura and I have had this debate a few times now. It's a debate about where do new ideas come from. It's about theories. It's about imagination. The last time that we were supposed to have the debate was in SRCD, which is a child development conference. But Laura couldn't make it. So I ended up pulling a Monty Python man wrestles with himself routine where I was arguing both my point and Laura's point. So it's nice to have a sparring partner again after that experience. I am still a little intimidated. Just to give you a sense of the caliber of my opponent-- the reason Laura didn't show up to our SRCD debate was that she was giving a talk of her own in something called TED. I don't know if you've heard of SRCD, but presumably you've heard of TED. OK. So what's this debate about? What am I talking about? Here's a 65 second prologue sort of setting it up. And then we'll get into it. The background to this debate is that, you know, we have this sort of really nice picture coming out of development and models. And sort of Laura and Josh have set up the perfect intro to that, which is from development we have this idea of children are sort of like scientists or hackers or what have you. They come up with these theories to explain the world. And they do it beautifully. And that's what they do in development. And then we have this idea coming out of computational land which is, well, what we mean by theories-- how do we actually capture that computationally? How do we formalize that? Well, maybe it's something like hierarchical learning over space of programs, right? Because programs are this the thing that's kind of like theories. And they kind of explain the world. And you can learn them via Bayesian inference and programs over programs and learning over learning. And that's sort of a really nice picture. But a lot of people-- Josh I think was using the phrase-- or Nour was using the phrase-- "give us grief." A lot of people give grief to these computation models correctly. They say something like, wait a minute. These computational models that you've described, like Josh was describing earlier, these hierarchical programs over programs, what's the theory space on that, right? Like, what's the space of all possible programs? Remind me. Isn't that infinite? And isn't infinite really, really big? What are you trying to suggest, that children are just searching through this giant space of programs? How could they possibly be doing that? And what we'd like to argue and I'll spell out is that while we don't search through the space of all programs, we don't do that when we write out our programs, right? We somehow figure it out. We're writing out these infinite spaces. And yet we search them in our computers. That's what we're try to suggest. The same way that we search them in our computers might be something like the way the children actually have to search for their theories. And it's hard, grueling work for computers and children. And the algorithms that we're, in particular, interested in-- there's all sorts of learning algorithms, of course. But one of the most influential learning algorithms that people have used for learning these programs and hierarchical programs over programs and things like that is just sort of this version of stochastic search. And I've told you a little bit about that-- those of you who went to the church tutorial-- about Markov Chain Monte Carlo, and all that stuff. But the point is something-- well, the way to search large, complicated theory spaces is through something like stochastic search. And, well, if it's something like stochastic search algorithms for our theories, maybe children are doing something like stochastic search, right? No is what Laura's going to say. Or she is going to say that can't possibly be the whole story, and here's what's missing. So does everyone sort of understand what the debate is going to be about more or less? I'm going to spell out, of course, all of this in the next few minutes. OK. So as I said, we're going to switch back and forth, Laura and I. The outline of the debate is that I'm going to give some background for like what good our theory is or presenting a good theory-- just really sort of covered that. So I'll be zooming through it just to give you an example of what I'm talking about. Because I want to tell you what stochastic search in theory space looks like. After I do that, Laura is going to step in. And she's going to rudely interrupt me and say, no, no, no. That can't possibly be true. Here's all the things that are wrong with that. I will then give a rebuttal. And then Laura will end and summarize. Beyond all these things, something I didn't write was all of you. Jump in and say what you think. OK. So what good is a theory? Like I said, Josh sort of went through this. I don't think you guys need that much convincing at this point. But by theory, I mean some sort of structured knowledge that goes beyond the data, compresses the data in some way, and is able to predict new things. My running example is going to be much simpler than character recognition or anything like that. It's going to be about magnets and, in particular, very simplified theory of magnetism. So suppose we bring a child into the lab. And we tell her, look at these blocks. They all look the same. Play with them. See if you can figure out what's going on here. OK? And unbeknownst to the child, but beknownst to you, is that these things are magnets. They're not just blocks. Some of them are metal. Some of them are magnets. Some of them are plastic. OK? So she's going to start playing with them. And she's going to start noticing something. Like, sometimes none of these things do anything, right? They don't stick. But sometimes, huh, they do stick. And she starts collecting observations in something like this. Like, how could she explain the data? Well, she could explain the data in this. You can just write it down. She could like label these somehow. She could have all these, A to I. And she could say, well, A and B attract one another, B and A attract one another, B and C attract one another, and so on. OK? That's her theory. That's her explanation of the data. This is horrible, of course, right? Like, this is not an explanation in any sense of the word. It's just a table of data. But in one sense, why is it not a theory? It's because it's just writing down what you've already seen. It doesn't compress it in any way. And it can't predict anything new. If I now give you a new block X and I tell you, listen, X attracts B, what else can you tell me about it? And she'll say, well, X attracts B. Yeah. OK. That's what my table tells me. You can't really predict anything new from a giant table like that. You want some sort of compression. You want some sort of theory. But what else could she come up with, right? There's all sorts of things you could come up with. But she could have come up with a very simplified theory that goes like this. Suppose that we imagine that there are certain things in the world. We're just going to call them, like, shmagnet and shmetal. Because I don't want to confuse it with actual magnets and actual metals. But she comes up with a name. She says there are several things in the world. And how do these things interact with one another? Well, let's just hypothesize some rules. And we can talk about how does she come up with these rules. How does she know that there's two things in the world? Let's say for some reason by dumb luck she hits upon this, which is actually a really good theory for explaining that, which is to say there are two things in the world. And the way they interact is through these rules. If something is a magnet and another thing is a magnet, if X is a magnet and Y is the magnet, they're going to interact. They're going to stick. If X is a magnet and Y in the metal they're going to stick. And interactions are symmetric. OK. So metals don't stick to metals, but metals stick to magnets. Magnets sticks to metals. And the interactions are symmetric. And if you have these rules, you need to pay some overhead, obviously, for remembering the rules. But you can compress this, you know, n by n thing into just the vectors of remembering what are the magnets, what are the magnets. So you've achieved some compression. And if I give you something new and I tell you, look, this is X, X attracts A, and you're like, wait a minute. A is a magnet. So this is either a magnet or a metal. You can probably predict a lot of different things about this thing. You can go and design your little experiment to figure out if this new thing is itself a magnet or a metal. Great. So this is wonderful. I won't go through something. Like, there's this added thing of finding out which one are the actual shmagnets and shmetals. And then you can predict the observed data. OK. So now, what we've done is we've set up this sort of space of possible theories, right? Imagine like all the possible logical theories that you could have written out, of all the possible logical predicates, there's an infinite amount of them. But now we've turned this into a rational inference problem, right? The problem for you as the learner is out of the space of all possible theories, just find the one that best explains the observed data, right? Where, by best we mean something like Bayes' law, right? Try to find the best theory to predict the data, whereby we mean the theory that's sort of the shortest and most compressed in itself a priori. You've all seen Bayes' rule. You've all heard Josh talk about this and other people. But it also explains the data itself the best. That's the problem. Go figure it out. We formalize it for you. We've solved it. And, of course, you know, this elides a lot of things. It doesn't really solve anything. But we'll get to that in a second. So you might wonder like how do we build a good space for possible theories. And one way to do then is to say, well, how do you generate all possible theories? How do you define a space of all possible theories? In this case, we're just going to go with grammar. But you could imagine something like a generator for all possible programs, right? So a grammar, to put it very, very, very shortly, is something that you can run forward and will generate a sentence. Or it's a way of sort of looking at the sentence and figuring out what its underlying rules are. Let's put it this way, OK? So in a grammar, you would start with a sort of a sentence node. And then you know that a sentence can go into several other nodes in some probability. For example, a sentence node can go into something like a verb phrase and a noun phrase. The noun phrase goes into a noun and another thing. And you sort generate through this grammar until you end up with a sentence. And it can be any sort of sentence. It can be the clownfish ascended the stairs darkly, right? And you can run this grammar forward and generate whole new sentences that you've never thought of it before. And you could then also use that grammar to figure out sentences. So if you see the sentence, the clownfish ascended the stairs darkly, you can use something like a grammar to figure out, well, you know, how is this sentence constructed. This is a grammar for logical predicates. It just means that if we run it forward, it starts out with some sort of abstract thing like, you know, it's going to be a law or some sort of set of logical predicates. You run it forward, and it ends up not with a sentence, like the clownfish blah, blah, blah. It ends up with a particular law that's relating a few particular predicates like, you know-- sorry-- like X and Y interact symmetrically, right? Like, there are things in the world. There are X and Y. If X interacts with Y, Y interacts with X. That's a law. We run through the grammar, we generate that. We could run through the grammar again, and it will say something completely different. Like, of all the things that I've related, there are such things as magnets. There are metals. And they should interact in this way or that way. Once we define a grammar like that, you can predict all sorts of different theories, right? I gave simplified magnetism, but you could also capture kinship with a set of four laws and several predicates. You could capture taxonomy. You could capture very, very simplified psychology of preference. OK. So as I said, this is another view of theory search. I haven't yet gotten to the algorithmic level of how do you actually search for these things. But I did want to set up the grammar, because it's going to be a little bit important later on. The grammar is just a way of saying, you know, you start with something. You generate it forward until you end up with a particular set of logical rules. So, again, to phrase the problem in a particular way of logical inference is to say we have the space of theories. It's an infinite space, right? And before I've seen any data, there is some probability of the right theory is to explain my data, where the data can be something like-- play with these blogs, figure them out. Before I've even seen the blocks, before I've seen anything that I need to explain yet, I have some sort of space of all possible theories. It can be all the logical theories, all the possible programs in the world. And I have some sort of probability distribution over which one of these are more likely than others, which comes from the prior. And the prior can be something on simplicity, like shorter programs are more likely, or the less free parameters the better. Something like that, right? So before I've seen any data, I can already score some programs as being unlikely, because they're too long and too ungainly. Does that make sense to everyone? OK. And that's just a representation of that in 2D space. It's saying, like, over here is theory space. And the size of the hills over 2D space is how likely each point, each point in theory space, is a theory, where theory can be, as I said, a set of logical laws or program or anything. And the height of the hill over that is the amount of probability you should assign to that theory, how much you should believe in it. And as data comes in, what happens is you become more or less certain of certain programs. Like, suddenly you shift probability distributions around. You say, oh, these things that I didn't think were that likely, well, the data's really pushing me to accepting these theories. OK? And that's what learning is, right? That's the view of learning. You just start with particular probabilities. You get some data. And then you shift that around, and you get other probabilities. And that's sort of a very beautiful picture. And the only problem with it is that it is clearly false. And the reason it's false is because this is an infinite space. And there's no way that you could have scored the entire infinite space. There's no way that what children are doing, what adults are doing, what computers are doing is to instantly shift probability mass over this entire space, right? New data comes in. They're holding in this infinite space and shifting exactly the probabilities around, right? This view of learning is ridiculous, because, well, it's patently absurd. And also, people have taken issue with it even though it was never meant to be the story of learning that is happening in the real-- well, how should I put this, in the real world? Do you guys know the difference between the computational level and the algorithmic level when I say something like that? Show of hands-- who knows about Marr's three levels? OK. Who doesn't know about Marr's three levels of explanation? OK. The point is to say when you try to explain the phenomenon, you're going to give it several levels of explanation. You're going to give it sort of the functional level, what we might call the computational level. And then you're going to actually say how does this actually get implemented in an actual machine. And there are many ways of implementing things. Like you might say, well, look, the general problem for vision is this, or this thing needs to be in addition function. These are the sort of things I wanted to do. But there are many different ways to implement that function. Some of them are better than others. That's the algorithmic level. How does this actually get implemented in an algorithm? Then you can implement the algorithm in many different mechanistic ways. You can go and implement it in neurons or in silicon chips or things like that. There are many different ways of implementing a particular algorithm. This view of, you know, you have some theories, you have some prior over theories, you shift the probabilities around as data comes in-- that's an explanation on the computational level. That's not an explanation of how we actually shift them around, how we actually search for these theories. And really the computational level sometimes gets grief, because people say, well, what are you saying? Are you saying that Einstein somehow had the theory of relativity, we all had the theory of relativity, in our head? And his process of discovery was just to say, well, I believe in Newton's theory. And I had some low prior on relativity. But then the data came in, and I shifted my probability to the theory of relativity. That sounds not the way people actually learn. That doesn't sound like the way we discover things. That doesn't sound like the way we come up with new theories. That doesn't sound like the way that children learn. And as I said, that's not exactly what we think. And that's not what happens in computers either. That's not the algorithmic level. So what happens in the algorithm level is you actually have to search for your theories, OK? And this is what happens in stochastic search in particular. What how does in stochastic search-- those of you who are in tutorial remember this-- you have some space of theories. OK, I'm still giving you the space of possible theories. Each dot in theory space is, let's say, a program. Or in this case, it's something like a theory in the sense I defined it before. It's a set of logical laws relating these predicates together. OK. So let me use-- I'm not sure you can follow my hand like this, right-- gaze detection or whatever. So theory A is, let's say, a theory that has three possible laws. It says if X has a P, P can be anything. And Y is a P. It's the same sort of thing. Then they're going to interact. The second law says if X is P and Y is a Q, they will interact. It doesn't matter. You don't have to figure this theory out, right? I'm just trying to give you an example of what a theory could be theoretically. And they interact symmetrically. That's one dot. Let's call that dot number one. That's theory A. Here is dot number two, theory B. OK? You as the learner, as the stochastic search learner in the algorithmic level, don't have access to the full theory space. All you have is A. That's where you are right now. That's all you have of the world. OK. That's how you can try to explain the world. And what you can do is you can try proposing certain changes to your theory. You can try taking out a law or putting in a law or taking out a predicate, somehow sort of messing with your theory, hacking with your theory somehow, coming up with a new theory that gives you theory B. It's a different point in theory space. And what you do then is you compare your two theories. You say, how well does this predict the data? So and so, you give it some score. You say, how well does this theory predict the data? So and so, you give it some score. You say, how likely is the theory a priori? How short is it? How simple is it? OK. How short is this theory a priori? OK. So you get some score for this theory that's based on the data in the prior. You get some score for this theory based on the data in the prior. And you basically decide which one of these two theories to accept. You could either stay with your old theory before you proposed any changes. Or you could decide that, wait a minute, that theory that I just proposed, the new one, is actually a bit better. Or even if it's not better, it's not doing that much worse, so I'll jump to it. OK. That's the stochastic part in stochastic search or in like these Metropolis-Hastings algorithm. This way of sort of jumping around in the space is the stochastic search I'm talking about. You're here in theory space. You have one theory. You propose a change to that theory. And we'll get in a second to how you propose a change to that theory. You propose a change. On end up here. You say, should I move here? Let's check. Is this theory doing any better? If it's doing better, move there. If it's doing worse, well, maybe still move there depending on how much worse it's doing. And this process is sort of jumping around in theory space. Probabilistically proposing theories and accepting them is not that far from what MCMC is doing or optimization through MCMC. Oh, sorry. I'm clicking on this. OK. And you can notice that this sort of search is somewhat different from-- this is the picture that Josh was showing before-- gradient descent or convex optimization, the sort of thing that you might do neural networks. I'm not trying to suggest that this is exactly what happens in neural networks. I mean, the more complicated neural network stuff, the energy landscape, can be quite complicated. And they still need to do stochastic gradient descent. But it doesn't look as fully connected and as horrible as these sort of theories on top look like. Because the space there is much more-- I don't want to say well-defined. They're both well-defined. But it's sort of much easier to search through, right? It's sort of easy to get in these neural networks to know where you're going to sort of differentiate and quickly get to your target. You're not so much doing this sort of hard laborious stochastic search. You sort of have this notion of, well, it's immediately going to be the best direction to go in is this. OK? I'm just going to do gradient decent. I'm going to roll downhill. And then that's some sort of good point in neural network land. OK. So how do we actually propose alternative theories? Well, you say, look, I have some sort of, let's say-- let's do this. OK. You have some sort of particular rule. And then what you do-- remember when I said you have some sort of grammar over theories, kind of like a grammar in language for those of you who know? The grammar describes a particular tree that you walk through to end up with your theory. What you do to propose a change to it-- and this is true for both these logical theories, but also for programs. You go to any [INAUDIBLE]. You go to any sort of node in that tree that generated your program or generated your theory, and you change it. You sort of re-sample from it or, you know, cut it and regenerate. And then what that ends up doing is it ends up, for example, adding a new rule or, for example, changing the predicate, or adding a new rule again, or deleting a rule, or deleting a predicate, or changing a predicate, deleting a predicate by sort of changing predicates, deleting rules, adding rules, adding predicates. I wasn't covering the full space. You jump around in theory land. OK. And this sort of dynamic of moving around in theory land stochastically-- one step forward, two steps back, you know, you don't know the target ahead of time, when you actually propose something, you proposed something new-- different learners might take different paths to get to the same target ultimately starting from an equal state of ignorance. You sort of propose different theories. You end up in different ways. But usually, if you get the right data, eventually you end up in the same spot. These discrete moments of sort of, you know, faffing about and sort of saying, well, this rule, that rule, that doesn't explain it. This explains it, doesn't explain it. Suddenly, you propose something that sort of clicks into place. You say, ah, it's not that there's two things in the world. There's three things in the world, right? It's not just metals and plastics. There's actually metals and magnets and plastics. Ah. And suddenly, you rearrange the data and things like that, right? All these things, all these dynamics, have something of the flavor of children's learning. And I didn't give you the full spiel. And you can go and read some of our papers. There's also been other people who have been very interested in sort of looking at the dynamics of stochastic search, how well it predicts what children are doing. Among them, I'll just mention people like Liz Bonawitz and Stephanie Denison, and also Tom Griffiths, and finally Alison Gopnik. So I also have TED speaker on my side. And not just any TED speaker-- it's Laura's former advisor. So at that point, it's a good point to hand it off to Laura and see what she has to say about this. But a midpoint summary is to say theories are useful, right? I think Josh has already convinced you of that. We want theories. The problem is that they define these rich structured complicated landscapes much more rich, much more complicated than anything that you might find in the neural network. Well, yes. And it's hard to search for these rich, complicated, fully connected landscapes, like the space of all programs or the space of all possible logic theories. And the way to sort through it is stochastic search, which can be horribly slow and wrong and things like that. But the claim is something like, what are you going to do, right? I mean, we want these rich theories. Rich theories define rich landscapes. And you just have to get away right now with stochastic search. Our algorithm solution for these spaces are stochastic search in that rich landscape. And, well, why shouldn't that apply to what children are doing? So here with a why not is Laura. LAURA SCHULZ: I think I have one. I'm hooked up, right? OK. So when I first engaged in this debate with Tomer, I was stuck on this kind of a thought that what's going to happen here is-- in which, following an eloquent exposition of a former model, attendant experiments, and quantitative data, Laura proceeds to wave her hands around. Someone was asking about intuitions. And I just had this strong compelling intuition this was completely wrong. But mainly, I was puzzled as to why I was stuck on this archaic locution. What was I doing with this "in which" situation? And it occurred to me the reason I was stuck on this particular locution is that I was thinking of a particular story that nicely illustrates exactly what I think the problem is with stochastic search. It's from a classic in developmental literature, the stories, of course, A. A. Milne and Winnie the Pooh. And this is the particular problem of "in which" Christopher Robin and Pooh and all go on an expedition to the North Pole. And so they're organizing a search. And like Tomer's algorithms, they don't actually know where the North Pole is. And it turns out, as Christopher Robin confides in a moment to rabbit, they also don't know what the North Pole is, also like Tomer's algorithms. But they do something about the terrain. And they know something about search. You gather all the friends and relations. And you engage in an iterative process over and over and over again until you succeed, and you find yourself at the North Pole. And Eeyore is a little bit skeptical about this. He says, well, you can call this you know, expo-whatever or gathering nuts in May. It's all the same to him. And at the of the day, I actually think Christopher Robin and colleagues here have a certain set of advantages over Tomer and colleagues. And that is that it is a Hundred Acre Wood. And so if the North Pole's really there, and they engage in that process, and they have all of rabbits friends and relations, they probably will find it. But it also turns out that, unlike Tomer and colleagues, they do know something about what they're searching for. Now, they're wrong about this. But nonetheless, I think they access an important constraint on the search process and actually helps them out quite a lot. So that's what I'm going to try to talk to you about today. So here are the issues with stochastic search. I think there are two big ones. Grant everything-- grant grammar, grant prior knowledge, grant templates, grant a bias towards simplicity. The problem with an infinite search space is infinite is a very, very, very big space to search. It's a very big space to search. And children seem to do remarkably well with it. So I'm going to, I guess, give you a toy fictional example. I'm going to give you a non-fiction example from my child. We were riding on an airplane. She was about 3 and 1/2 at the time. And she knows a lot about airplanes. She has a lot of folk physics, prior knowledge about airplanes. And she also knows a lot about phones. She's had a lot of experience of phones. But nothing in her prior knowledge predicted the announcement that you have to turn off your cell phone when you fly a plane, right? So this was surprising. And she immediately said, well, I know the answer. I know why you have to do that. And I know that she doesn't know anything about radio transmission or government bureaucracy. So I said, how do you know? Why do you think? And she said, well, because when the plane takes off, it's too noisy to hear. You know, that example is not especially clever or especially adorable. But what is really, really interesting about that answer is that although it is wrong-- it's even wrong as to the causal direction-- it is a good wrong answer compared to all of the other things that are consistent with her prior knowledge and the grammar of her intuitive theories that she didn't say. She didn't say, because airplanes are made of metal and so are phones. Because airplanes fly over the Earth, and the Earth has phones. Because airplanes are big and phones are small. Because airplanes-- her grandfather lives in Ohio and has led her to believe that everything is made in Ohio. Because airplanes and phones are both made in Ohio. Infinite is a very big space. There are a lot of things that you could say consistent with prior knowledge where you're making random changes in your intuitive theories that are not even wrong. They're not even wrong. And so the real question is, how did she converge at a good wrong answer, at an answer that, although it is wrong, makes sense? It isn't just, I think, a toy problem. But here is the problem. There are innumerable logical constitutive causal and relational hypotheses consistent with the grammar of intuitive theories. How do we so rapidly, literally between the announcement and the next thing out of our mouths, converge on ones that, if they were true, might explain the data? They might not be true. But if they were, they could work. They could solve problems. And that I think is the really hard mystery. And again, it's not just a toy problem. Modeling even relatively simple well-understood problems takes time. I would often come across Josh's students wandering in the hallways. And I say, what are you doing? They're like, oh, I'm waiting for my model to run. I'm like, waiting for your model? Computers are really fast. They're fast information processing. What is it doing? Well, it turns out it's generating a lot of answers that aren't even wrong. That's what it's doing, right? It's spending a lot of time sitting around sifting through things that aren't even right. Iterations are spent searching in hopeless places. And this is true of some of the best and the brightest. This was like a fantastic NIPS paper, a major advance. It's a probabilistic graphics program. And it's solving the really deep theory problem of CAPTCHAs. So it's trying to figure out in this case-- but it doesn't just do-- in fairness-- well, let me return to that. Let me show you what it does do. There's a rectangle over there. It wants to be able to figure out what is in that space. And it wants to model it. It wants to find a rectangle in the lower left-hand corner of a scene. So what does it do? It generates pixels all over the map until it finds the rectangle. And I saw this in [INAUDIBLE]. And I said, why is it looking all-- can it at least confine its search to the lower left-hand corner? But, of course, the algorithm doesn't know from lower left-hand corners. It's a powerful algorithm. If you wanted to solve a CAPTCHA, you could do-- see look at it. It's all over the place. Now, this is a virtue. And, yeah, it converges. And that's just great, right? But why doesn't it search in lower left-hand corner? Well, the answer is it doesn't know from lower left-hand corners. And it's a feature, not a bug, that it doesn't know from this. Because that means it doesn't just solve CAPTCHAs, which you can do with edge detection. It can find, you know, objects in the road. And it can do all kinds of other things that I'm sure Josh and Vikash would be happy to talk to you about. It's a pretty general thing. But it's not constrained. And that's a feature. But the interesting thing about humans, including human children, is they are both flexible and constrained. They can both solve a whole bunch of problems, and they can converge on them very quickly, right? Whereas, here, it trades off. And, again, that is a simpler problem. That is a square and a bunch of pixels. The kind of learning we're talking about when we're trying to talk about theory generation or real world learning is a really, really, really, really big space. So one problem is just how are you going to get at least to answers that, if they were true, might work? But if one problem is that the theory space is really big, the other problem is that human learners are not that dumb. We have a lot of knowledge. And we have a lot of knowledge that these algorithms are not making use of. And the question is, why not? And is there any way that we could develop models that did make use of it? In particular, we know a lot about our problems. Our problems are actually our friends. We know about our problems and our goals. And we know about our problems well before we can solve those problems. An abstract representation of what the solution might look like, what it ought to do, what the criteria it's trying to satisfy are, could help constrain and guide the search. It matters about it though, not just that she had prior knowledge about airplanes and about phones, but that she had prior knowledge that the problem she was solving was an unpredicted incompatibility between airplanes and phones. You have to turn one off when the other is going on. That's information that's not in her general background knowledge. It's about the particular problem that she has. And the question is how could you use that to make good proposals and make better proposals? So the proposal I have is that when you know something about what you're looking for, then that can help you find it. And this is the kind of knowledge that Christopher Robin and colleagues had that Tomer and colleagues did not. They at least eventually decided that what they were looking for was, of course, a pole. They know what poles are. Therefore, when they find a pole, they can be quite confident that here they are. And this is a good candidate solution to their problem. That solution is wrong, but at least it's wrong. OK. So the argument here, which I'm going to try to get slightly more precise, is that the form of the problem as an input to the algorithm should increase the probability that proposes useful ideas. And you can consider this even in the simplest form in the kind of information that is contained in our question words. So I think it's an interesting feature about human cognition that we have a very, very small handful of question words, which we use to query the entire universe. And you know what? Those question words do a lot of work for us. When I tell you I'm asking a question about who, you might propose that we ought to be looking for some kind of answer that's something like a social network. And a social network might be more likely as an answer than a 2D map. Whereas, if I ask you a question about where, well, you really do want to consider 2D maps. If I'm asking when, you're talking about a time line answer. If I ask you a why question, maybe it's a causal network. And if I'm asking you a how question, maybe it's a circuit diagram, right? You don't know anything about the content at this point. I could be asking you anything. But I ask you who was Christopher Columbus, and you answer 1492. That's the kind of thing that our algorithms are doing. That's not even wrong, right? It's consistent with your prior knowledge. And it's the kind of thing Watson does as good [INAUDIBLE] solutions. But it's not what children do. It's not what children do. I think that the issue is that this is actually a friendly amendment, right? Because in what we have shown time and again-- by we I mean not me, but computational modeling folks-- is that we can use lots of information out there for hypothesis evaluation, right? Once we have a theory and we have the data, we can select and use this information and say, well, does it answer the problem or not, right? Does it improve? Does it make better predictions? So we use this kind of information that we have. Even formally, we can say that we can use it to select among hypotheses. We can use information about the structural form of problem to represent them. The question is, can we use the same kind of information to constrain the search space? And it's easy for me to say. I don't have to do that, right? But it's the kind of proposal that I think is missing. Because we have rich constraints that go far, far, far beyond our question words. The kinds of problems that we have and the criteria for solving them derive from all kinds of sources. We try to solve different kinds of problems-- navigation in some cases, explanation in other cases. And some of those are epistemic ends. I want to persuade you of something. I want to instruct you in something. I want to deceive you in some ways. But we also have all kinds of non-epistemic goals. I want to impress you. I want to soothe you. I want to entertain you. Each of these goals is actually a constraint on what is going to count as the solution. Our goals are innumerable. But there are only a small handful of ways you can solve any given goal. So when you're dealing with an infinite search space, having a goal, having a problem, actually could act as a constraint on how you search for the solution. And it is an interesting feature of human cognition that our goals can be noble or venial. They can be impressive or trivial. And it may not matter with respect to the solution we have. We have analytic logic, because the medieval monks wanted to find incontrovertible proof for the existence of God. We don't hold onto their goal, but we hold onto their solution. We are here in the East Coast of Massachusetts, because of the search for the West Indies, right? So our goals act as constraints on the solution and on the search process. And the importance of our goals may be that they do exactly that, that they help leverage some new search in a way that at least helps us make progress. So the argument here is instead of stochastic search, that we have-- I don't call it goal oriented. I call it goal constrained now-- goal-constrained hypothesis generation. And the idea here is that at least we know something about where we want to go. Now, this is not a total argument against stochastic search. It's just a way of getting stochastic search into a much smaller search space, right? Once you know what things count, then you can do everything Tomer says you do. I actually agree with him. But you don't want to do it over all possibilities, because you know a lot more than that, right? You know what kinds of things are going to count. You should do it over that space, not over-- should look in the left-hand corner. Then you can iterate all the pixels you want. But if the thing's on the left-hand corner, that's where you ought to be looking. I'm going to give you a corollary to this, which is if you don't have any idea what the search space is, you are going to resort to an extremely inefficient, extremely frustrating search and, actually, the kinds of conditions under which I think human beings quit. So I will give you an example from my personal experience, as Jessica Sommerville knows. Because we were trying to get my child's booster seat-- because children now need booster seats until they're 14-- re-attached from my plane flight. Couldn't do it, right? One thing goes down, two things go up. It's a spatial problem. We spent 10 minutes, two PhDs, on it-- threw the thing out and just had her ride in the bottom of the booster seat without the top. If I have 1,000 piece puzzle and I have to find a puzzle piece, people who are good at puzzles know before they find that piece something about what it's going to look like. Like, OK, well, the edge has to be angled like this. It has to have a concavity here and a convexity there. And that's what I'm looking for. Me-- soon as I look away from the piece, I have no idea what I'm looking for. And so I do what Tomer would do. I do a stochastic search over all the puzzles, And with 1,000 pieces and many permutations, that is not a good way to solve a puzzle. As a result, I never do puzzles, right? As entertainment, I just don't understand. But if you do know, it's very satisfying, right? You know what you're looking for. And now you can constrain your search space much more effectively to those kinds of things. Indeed, when I say we are smarter than that, human beings know. We have metacognitive principles around these kinds of things, right? This knowledge is in human minds, what it might mean for us to think that a problem is tractable. What does that word mean, right? Sometimes it means we have the financial resources or the technology to carry it off. But often, it means we have a well-posed problem. We don't know the answer, but we know what the answer needs to do. We know what the answer needs to look like. We have criteria for what would count. At least we have a precise enough representation of the problem to effectively and efficiently guide the search. And I think that's the kind of thing we would ideally like our models and algorithms to have. At that point, we may have to bounce around a lot. But we're bouncing in a pretty well-defined space. So to the degree that this is true, it actually explains a lot of otherwise peculiar features of human cognition. For instance, we had this weird sense that we're on the right track, right? Well, what does it mean to be on the right track? It surely doesn't mean you're better at explaining the data, right? You may be nowhere close to having an answer that makes better predictions or gets it right. But you're like, oh, that's a really good idea, you know? You get excited. Or you're like, no, that's a non-answer. What does it mean? It might mean that at least it fits the abstract form of solution. If it were true, it might work. We can tell our students in an undergraduate class that that was a great idea even when we know it's been disproven, right? So it's actually false. It doesn't explain. It's just not true. We still think it's a great idea in virtue of what? Not it's fit to the data, right? Not how well it's predicting things. It's actually false and still good. What could that possibly mean, right? It must mean there's some other constraint on hypothesis generation that we are sensitive to. So I want to suggest that there are actually two constraints for our hypotheses. One is how well they fit prior knowledge and data. That's the one we know something about. That is, for instance, truth. But Stephen Colbert in his infinite wisdom proposed something else. He said, well, we're also sensitive to this thing called truthiness, right? You know, like how good the story sounds, how good the argument. And of course, in politics, you know, he makes massive and effective fun of this. But I think this is a feature of human cognition, not a bug. I think it is extremely important that we can generate ideas that are truthy, right? Because they're plausible. They're interesting. They're informative. They tell good story. They may be false, but it could be worse than false. It could be not even false. So I want to make an important point here. Generating new ideas is not just about Einstein versus Newton. And it's not about going from an undifferentiated concept of heat and temperature to a modern scientific one. It's just about radical conceptual change. This is the stuff of ordinary everyday thought. It is our ability to reliably make up new relevant answers to basically any ad hoc question. The answers may be trivial. They may be false. But they are genuinely new. And that we didn't have them until we thought of them, they're genuinely made up. We didn't learn them from evidence or testimony. And they answer the question. They're not non-sequiturs. And I think this is important. And this is only possible if we can use the form of our problems to guide search. So let me give you a few examples. What's a good name for a theater company? None of you know. You haven't thought about that problem before. But now you're thinking about it. Well, what's a good name for a theater company? How do you get stripes on peppermints? This is not a problem you walked in thinking about. None of you are working on this as your independent project. But you can think about it. And you already know enough, knowing nothing about theater and nothing about peppermints maybe, to know what the constraints, what counts, right? That's the kind of information you already have. It's not prior knowledge about theater companies or peppermints. It's part knowledge about what's going to work for a solution. So for instance, you know that McDonald's is a nonstarter. And [INAUDIBLE] or whatever is also not a good name for a theater company. You know that getting strips on peppermints you don't want to do it with a spatter [INAUDIBLE].. You don't want to just spray things at it, right? Because that wouldn't even count as a solution. That's not the kind of thing you're looking for. You're looking for something more like fresh ink. It's new. It's novel. It's familiar. It makes some reference. You're looking for something more like a pendulum approach, which at least could generate periodicity. I'm sure they don't use a pendulum to spray paint peppermints. So are you. But it's not a bad answer, right? So is there any evidence that kids can do this, that information contained in only that strict form of the problem can help learners converge on solutions? We wanted to find out. So I'm going to show a little baby attempt to start getting at this problem. We gave kids a machine. And we gave kids some things that could work the machine. And the machine had two visual effects. You could make a ball appear to flow up and down on that screen there. Or you could make the ball appear at the bottom and then flash up at the top, right? Because we put a computer behind, right? So woo, the ball is moving continuously, or the ball is moving discretely. And the affordance, as you might note, are also continuous or discrete, right? There's a rolly ball, or there's this peg that you can move back and forth continuously. Or there's a peg you can pull in and out, or a drawer you can pull in and out. So there's continuous and discrete affordances and also continuous and discrete effects. So we also had an auditory tone that varied continuously and an auditory tone that went from high to low. Everyone got it? And we just showed the children all the parts that connected to the machine. And then we showed the kids the effects, but we had hid the affordances. And we said, well, which part made the ball go? And we asked them either about the visual or auditory affordance? They'd seen no covariation data, prior knowledge- you know, agnostic about all of this. And the question is, well, would they say, well, I'm trying to solve a problem about a continuous effect. I should use a continuous affordance. I'm trying to solve a problem about a discrete effect. I should use a discrete affordance. Would they use something about the form of the problem to answer the solution if they knew nothing else about it? So there's no fact of the matter here, right? Because, obviously, we're not using either of these really. But the prediction was that they would indeed make this kind of mapping. And they did so. Now, what you might worry about was that, well, there is no fact of the matter. So they're just doing some kind of cross-modal mapping, right? If you don't have a way to answer the problem, they're just saying something like, well, you know, this has this property, so does this. Let me go ahead and make a mapping from one to the other. It's a weird kind of cross-modal mapping. Because, usually, you integrate things that are in a single stimulus actually contained together, like the sound of a ball dropping and the sight of a ball dropping, not two different things. But maybe it's something kind of like that. So if they're actually using the form of a problem, then if you change the problems, they should generate different solutions. So what we did in the second experiment is we said-- oh. Yeah. So what we did here is we showed the children the continuous visual stimuli, and then we asked them to generate the continuous auditory stimuli. OK. So all of these are now continuous. They could still just go ahead and make the continuous map. But if you represent the problem as changing from visual to audition, that feels like a discrete problem, right? You're completely changing modalities. So now at this case, you might expect the kids to resist making just the continuous mapping if they're using something about the kind of problem they have to constrain how they search for the solution. Now, again, there's no real right answer here. But what you see is the kids shifted their responses in response to this. So this is just a tiny bit of suggestion that when there's no differentiating prior knowledge and there's no differentiating evidence, children take into account what kind of problem they're trying to solve, and what the information is, and the problem itself that can help constrain their search for a solution. I'm going to give you another example for some more recent work by Pedro Tsividis in our lab. He varied the dynamics of a scene. He had here some bugs. And in one scene, those bugs in green, they varied periodically. So they just went from having very few spots to having a whole lot of spots to having a very few spots to having a whole lot of spots, right? That's what these green bugs do. And the other bugs just got faster over time. So those longer vectors are supposed to be indicating the bugs are getting faster continuously. And then he said, well, you know, here are some bugs in these rooms. And I have two kinds of lights in the room. One set of lights looks like those on top. The other set of lights looks like those on the bottom. Can you tell me which lights are responsible for the behavior of which bugs? Again, a sort of very similar kind of problem. No fact to the matter that if children can represent something about the abstract form of the problem and use that to constrain their search for the solution, then the periodic lights should reflect the periodic change in the bug's behavior. The continuous light should reflect the continuous change in the bug's behavior. We are, of course, not using the words periodic or continuous with us or anything like that. They have to pull that out and say, this is the kind of problem it is. This is a feature of a possible solution. Let's go ahead and make that mapping. And indeed, the kids are doing that well above chance. And obviously, in this case, we're giving the kids some possible solutions. They're not generating it whole cloth. But minimally, it's a different way of thinking about problems and search. They're not using most of our sort of traditional ways of figuring it out. They're just using something about the kind of problem they have and what's available in the problem to help sort out the solution. So is this analogical reasoning, right? It feels kind of an analogy. But it's a funny kind of analogy. Because what it isn't is a mapping between a known problem and a known solution to a new problem and a new solution. It's rather a mapping from the kind of problem you have to the kind of solution you have. It's using, again, the problem or the query itself as your friend. It has information in it about how it wants to be solved. How are you going to use that to solve it? And, again, the virtue being that even if your answer is wrong, if it were true, it would work. So the argument also, of course, is that this applies to any possible goal we might have including those cases where it's just not at all obvious how an analogy would apply. So what is a good name for a theater company, right? You're using something about what would count as a solution to constraint something about what you think would be a good answer. But there's not an easy way to tell that story as analogical reasoning. So their argument is children seem to have data independent criteria for the evaluation of hypotheses. And these criteria extend beyond simplicity, grammaticality, or compatibility with prior knowledge. They consider the extent to which a hypothesis fulfills the abstract goals of a solution of a problem, not just the degree to which it fits the data. And I will suggest that this is maybe deeply part of an important mystery of human cognition, which is-- our most powerful learners spend a lot of their time doing something that has defied most of our best attempts to explain it, which is they've spent a lot of time, many hours a day, just pretending and making up stories. Stories have some interesting properties. They do not have to be true, right? They don't have to fit the world or fit the data. But they do have to set up a problem and a solution that, if true, would solve the problem. Most of play has that. It is not at all obvious why you would think it was important that you want to balance a twisty on the top of the candle stick in order to shoot pee through it. But it's a problem. And guess what? If you can set up the problem and find the solution, you've just accessed a really important ability about setting up problems and setting up solutions that, if they worked, would solve that kind of problem. And, of course, imagination, most of our narratives, most of our fictions, have at least those sets of properties. So I'm going to go ahead and stop there. [APPLAUSE] TOMER ULLMAN: OK. So if you remember the structure we said at the beginning, there will now be a short rebuttal. And then Laura will give an even shorter summary, and then the free for all. So most of you are probably thinking all sorts of ways that Laura is wrong. But wait, let me get through it first, and then see if I didn't cover something. But, actually, my response to this, to all of what Laura's said, is not you're wrong, but you're right. So you're right, Laura. You're right. Other people in the audience who think that stochastic search by itself-- if you have some sort of infinite theory space that was supposed to account for all possible problems and for any new problem what you did was to just completely at random try to search through that space anew, then that wouldn't work-- that's fine. And I agree that there's something inherently wrong about an algorithm that can take some problem, like why these two blocks are sticking together, and say, well, maybe it's because the moon is bigger than a piece of cheese, right? Like, as Laura said, it just seems like it's not even wrong. Or maybe it's because people have more than two children on average. No. And there's also something wrong. Laura didn't quite get to this or maybe not emphasize it. But she emphasizes it sometimes, which is there's something wrong about-- well, she did a little, but-- an algorithm that makes sort of dumb proposals. Dumb proposals of all sorts of things-- things like you try to explain something in theory space, and you say, well, maybe it's because of X. And you check it, and it's not X. And you say, oh well, oh well. Maybe it's because of X, right? There's something wrong about stochastic search. Although, I have to say, Laura, you have an eight-year-old. And, you know, when we gave this first, I had a two-year-old. And I actually think maybe it's X, maybe it's X, maybe it's X is not such a crazy way of describing what two-year-olds are doing. LAURA SCHULZ: Well, [INAUDIBLE]. So, you know, like, what if there's noise? TOMER ULLMAN: Yeah. Yeah, yeah. OK. But I do want to give an actual response. So, you know, I think I have some responses for Laura. And these are responses that, importantly, you know, they'll try to address what Laura is saying. They'll try to take it into account. They'll try to give new answers to it that will importantly leave her unhappy. So what I'm going to try to do is take a page out of Hannibal Barca at the Battle of Cannae and try to envelop. So I'm going to highlight of work by other people. From one direction is work by Steve Piantadosi, which is about making these, you know, algorithmic search part of the problem, whether it's too slow, students are wandering around the halls doing nothing waiting for it to converge. What if we just did it really, really fast? Another way to address this is to say, what if we made actually good proposals? OK. The problem of making proposals that are just ridiculous-- what if we made proposals that actually take the data into account a little bit or the previous data, the stuff that we're trying to explain? Another way to address what Laura is saying is to say, well, maybe we can make better primitives. And better primitives mean that your search space is actually more confined to the right sort of things. And finally, I'm going to highlight some new thoughts by myself about ad hoc spaces and how we might construct them to get at this problem of this thing, like how would you come up with a new name for theater company or a name for a new theater company? So I'll go through these somewhat fast. You're welcome to come and talk about any of them. And I'll point you to the people who are actually doing the work, which I've said before. So let's see. This is just sort of the rebuttal. One of the rebuttals is to say, well, here's this work by Steve Piantadosi, which is, you know, introspection is really actually a poor guide. So when [INAUDIBLE] was giving that beautiful example of cell phones, why you have to shut them off on airplanes, you say, oh, and she came up with this example that it's not true. But if it were true, it would be nice. Well, maybe she went through a billion other things before she came up with that? Laura's like, she did it like that. But we don't have a good sense for what is actually fast, what is actually slow. And there might be a case where we don't actually introspect about a lot of things. The things that bubble up into consciousness that you might actually accept or reject rely on actually a ton of other proposals that they don't even bubble up into introspection that you're making very quickly and just rejecting even before that. And Steve came up with this really interesting way. You guys have probably heard about deep learning and the sort of the GPU revolution for deep learning and things like that. So the point is if instead of we use CPUs we would use graphical processing units, then we can make stochastic search algorithms in parallel in some cases. And once you can make them in parallel, then you can put them on a GPU. And once you put them on a GPU, a GPU is sort of a way of taking something that's supposed to be a CPU. And instead of having a CPU that can do something sort of complicated on a sort of task, you can do a lot of really simple tasks in parallel very fast. So if you could make that proposal, that thing I said before-- take a theory, make a change to it-- if you could make that into the sort of thing that you could put on a GPU, then you can make a ton of those proposals very quickly in parallel. And that's sort of what he figured out how to do for a bunch of spaces. It's much faster than the CPU. And the main advantage is that it's also much cheaper. And you can cram a whole bunch of together. And you can get to something like-- I forget the exact numbers. You can make like a million of these theories proposals a second. And that's just with today's technology, right? We don't know what's coming around the corner a few years from now. You know, Steve plus GPUs is awesome. And you could think of it like these various problems like you're trying to fit these data points on the bottom. Can people see that? This is sort of a classic problem. You have some data points. You're trying to fit a polynomial to it. And you're trying to say, well, how will we do that? The truth is there's a lot of very clever ways of doing that. But let's assume that you're even doing random search in polynomial space-- not the sort of thing you want to do. Those of you who have been to the tutorial, I mentioned that if you have an actual better way than stochastic search, you should probably do that. But suppose you didn't know and you wanted to do stochastic search. You could still do a million moves a second and quickly converge on something like that line that you see. OK. And that line is actually taken from, I think, Galileo Galilei's data for how things slide on a hill. I'm not Galileo Galilei sat around and said, maybe it's x to the square, maybe it's x to the 2.1, maybe it's x to the 2.3, maybe it's x to the 2.1, maybe it's x to the 2, and then finally converged like after a million moves to that. That's not exactly scientific discovery. But for a lot of everyday thinking, you might actually be proposing things very fast and rejecting them. OK. That's Steve. Some things from Owen Lewis about making maybe smarter proposals-- and this gets at that point of, like, maybe it's an x, maybe not. So I suppose that I'm trying to teach you a concept. OK. I'm trying to teach you a particular concept. I'm going to give you some positive examples of the concept. OK? This is the sort of thing that psychologists really like to study. OK? So this is a room-- no, Roomba's an actual thing. Blick gets overused. Can someone give me a nonsense term? This is a Jabberwock. OK? This is a Jabberwock. I'm going to give you another example of a Jabberwock. Who thinks they know what Jabberwocks are? You have some sense of what a Jabberwock is? OK. Huh, that's also a Jabberwock. Wait a minute. OK. The sense that I had for maybe what a Jabberwock is is maybe not that great. That's a Jabberwock. That's a Jabberwock. That's a Jabberwock. That's a Jabberwock. OK. And you might think at this point, well, fine. Jabberwocks are either squares of any color or red circles. Or maybe they're squares or circles. I don't know exactly. You're building up some sort of theory for that concept, which can be described in something like a grammar for your current hypothesis. You might say it's either a red circle, or it's the square of any color. OK. And that's sort of your grammar for these concepts. And now you could sort of change that grammar, right? You could sort of excise these nodes and that tree to come up with new things. Why would you want to come up with new things? Because, look, that's a Jabberwock. OK. I just gave you a new example. It's something you didn't know before. It's sort of confounding with your theory. You have to come up with a new theory on the fly. OK? Theory-- again, used in a very, very minimal sense here. But if you accept that this is something like a theory, you have to come up with a new theory for explaining why that's a Jabberwock and all the other things that you've seen are Jabberwocks. What is a bad idea? A bad idea is to just cut and generate randomly, right? You might come up with something like it's a triangle or something like that. But you might come up with, well, maybe it's a square. No, we already did that. Well, maybe it's, you know, just triangles. Maybe it's just a square. Like, you could spend all this time not taking into account your previous theory and the fact that your new example had something to do with triangles, something to do with red triangles. You want to be able to make proposals that take into account this new data. Does everyone understand what the problem we're trying to get at is? So what Owen has done is to sort of take these stochastic search algorithms and say, if you get a new piece of data that contradicts, that sort of interferes with what you had before, how would you make proposals that must take into account this new data? I'm going to recut and generate. But I'm going to identify the places in my theory that would take into account this new piece of data. And I'm going to make smarter proposals. They might still be wrong. And there's better and worse ways of doing this. It's still going to be a randomized search and theory space. But it's at least going to take into account this new data. OK. And that's just-- I'm afraid I don't have time. But, look, it works. And it's much better than just bouncing around completely around random as you might guess. Another response, this work by Eyal Dechter. Let me skip over this for a little bit. This is work Eyal Dechter, which is to say, what if you wanted to use better and better primitives? OK. So before we have this notion of you're just bouncing around in theory space, you're making all sorts of notions, let me put it this way. Suppose that you're searching through the entire space of programs. OK? And the only thing that you had to work with is something like that lambda expression before, right? You didn't have the notion of plus, minus, multiplication, sine waves, things like that. And you're trying to figure out something about an equation. And you work through it. And there's a way of doing it. There's a way of, like, generating functions that rely on other functions that rely on other functions in the complicated way that will give you the plus function. OK? And you do that. And then you generate a lot. And you somehow manage to find out the sinus function. And you finally figure out that this function you're trying to describe is sinus x plus sinus y. Let's say something like that. OK? The exact example doesn't matter. But you worked really hard, and you figured that out. Now, you get a new example. And underneath the hood, it's actually just sinus x minus sinus y. Or let's say it's sinus y plus sinus z or something like that. And now you start all over again. You're like, fine. OK, lambda something, something, something. Like, if you could only use the fact that you've already discovered the sign function, you've already discovered plus and minus and things like that, and now when you come to try and explain a new problem, you actually have a lot of previous knowledge. OK? So when you're trying to describe why airplanes take off and how they do that, you're actually going to rely on previous knowledge. You're not going to search through your entire theory space starting from nowhere. You're going to rely on primitives before that have been useful. So you might notice that actually plus and minus and sines and things like that and cosine and exponentiation are really useful. Let's save those as primitives. So that next time that we make random proposals in theory space, you can think of it of making like a whole bunch of moves at once that were useful. OK. Like, a whole bunch of stuff at once that was useful that shows up all the time-- you want to make that again. OK? So I try that, for example. And the examples that we gave in theory space-- like you might find out that actually a lot of theories use transitivity, or a lot of theories use reflection, right? In the particular magnet case, if X attracts Y, then Y will attract X, in general, the law of if X blahs Y, then Y will blah X, that turns out to be useful in a lot of domains if only there was a way of reusing them, right? There is. And what Eyal was doing was to basically use an explanation compression algorithm, the EC algorithm. And what it does is it tries to encapsulate useful concepts. And he used it on a whole bunch of stuff. One of the nice domains he used it on was these circuit diagrams. Have any of you actually had to solve circuit diagrams? This is the sort of stuff that people at MIT do. You're given a particular input-output function. And again, like I said, under the hood it might be something like you're just told something like here's X and Y, OK? X and Y can each be 1 or 0. OK? And now I'm going to give you combinations of values of X and Y and what that spits out. OK. So X and Y are both 1-- light turns on. X and Y are both 0-- light turns on. X is 1, Y is 0-- light turns off. OK. And you're trying to find out some sort of circuit that will explain this behavior, some sort of combination logic gates, like ANDs and ORs and NANDs and things like that. OK? Do people sort of understand the problem for these circuits? And you can get a long list of things, X, Y, Z, T, some sort of complicated behavior. You might not even get the full behavior. And you're trying to find sort of the minimal set of logical predicates, or in this case circuits, that will explain that behavior. And now suppose that you only have the gate NAND to work with. Do people know what NAND is? OK. It's a sort of logic gate. It's a very simple logic gate that you can build up all the other logic gates from. But it will take you a while to build up AND from NAND or OR from NAND. But you can do it. And so what he did was, and his colleagues, they started out sort of giving this algorithm a lot of different problems. Like, here's a bunch of circuit diagrams that you need to solve or circuit problems that you need to find the diagram for. What the algorithm was doing was that each time it solved a problem or set of problems, it would go back and look, huh, which parts of this can I encapsulate, right? Which parts of this can I sort of use again? I can carve off a chunk of something that was useful. And now, when I make a new proposal, I'm not going to say put a NAND here or a NAND there. Stochastically, I'm going to sort of put in a whole chunk that I've already used before. OK? I'm going to sort of call that a new primitive. Cut out this part of the space. Under the hood, it's actually an AND or something like that. And discover-- so discover is not an AND. And discover is this really useful thing that they called E2, which doesn't appear in logic books. But it turns out to be really useful for certain diagrams, which is take an input, split it into two, do something on this part, do something on that part, and recombine it. It turns out to be a hugely useful concept for circuit diagrams. And this thing discovers it. And once it discovers it, it sort of reuses it. And what that does is it turns an infinite and unmanageable space into infinite, but a bit more manageable. OK? So your space is infinite. You're not going to search the full length of it. Imagine that this is a space of all possible programs. As you go down, the programs get way too long. You're never going to reach them. But some of them are really good. Some of them are really good explanations. And the only way to get to them is if you had some sort of way of chunking the problem, of saying, yes, it looks like a long program. But actually half of it I've already used before to solve a different thing. And half of it is less long than two times. So you might discover, you know, you might have an effective search area. You find out all the problems you can solve there. Yeah, this is an interesting thing, choice color. So imagine that of this blue thing over here is describing, within the space of all possible programs, the sort of programs that you want to find. So there's the effective search area. They only cover part of this blue thing. You can think of it like the probability is really high over there. You really want to find all of them. But by searching that small space and, within that small space, finding the right primitives and encapsulating them, you can now actually search more efficiently the rest of the space. And the rest of the space sort of compresses and compresses until it's all within your effective search area. Do people sort of understand that? It's sort of there were long programs before that you never would have gotten to. But by searching these small spaces that you could search through before, discovering the useful parts there, these new things that seem really long before are actually short. Because they can be described by just a few chunks. OK? This is a really interesting work. And I encourage you all to read it, those of you who find it interesting. The last thing I'm going to do-- and then that'll leaves 10 minutes to discussion, which is great-- is this problem that I guess maybe it's really the heart of what Laura is getting at. I think she was not satisfied by any of these things. And she was sort of pointing out, well, fine, you can do stochastic search all you want. But the really hard problem is constructing the space itself on the fly. You're not going to use one infinite space for all possible problems. You're going to use the right spaces for the right problem. And how do you do that? In this case, we're going to do, give me a good name for a romantic drama. All right. And your search space is going to be imagined that-- can people see sort of the border? Like, there's this whole space of uselessness. And what you really want to do is focus in on that tiny part of useful things. If only there was a way of just on the fly, you know, zooming in on that thing and then bouncing around in that. And the point is to say, well, when we construct the space, we can just use previous examples. I don't think it's the case that we just knew something necessarily completely new in these sort of everyday thinking. Well, maybe. We can argue about that in the discussion. What you actually start out with is actually taking a few examples that you find relevant in some way and using those examples to then construct your space on the fly, right? You might think about things like, what other romantic dramas do I remember in the past? What do they share in common? What movie names do I know of in the past-- quickly finding the sort of relevant thing for all these things, and then having the space for those, and then searching around stochastically. Because you're not going to do better than stochastic search. There will come a point where you're just bouncing around at random. So I used this actually, forgive me, a paper title for SRCD and came up with some amusing things. You guys can play with that online if you want. But let's do, give me a good name for a new romantic drama. So as I said, what you would do is you would just think about all the romantic dramas that you know, like The Climbers, Christine of The Big Tops, Cupid's-- these are all actual romantic dramas pulled off of Wikipedia-- then use those to construct your space. Don't care about all the things that could happen in the world. OK? And what do we mean construct your space? Well, there's a bunch of ways to look the space. What ideally we would want and what I didn't do, but what we're thinking of, is to construct a very, very simple grammar which instead of all possible sentences is a grammar for movie titles. And this grammar usually tends to generate things like the, right? The something something. And [AUDIO OUT] long. And then it just stops. OK? And it turns out that something like the adjective noun is a really good way of generating names for pubs-- The White Queen, The Blond Tiger, The Bleeding Bottle, I don't know, something. Right? That's really useful if only it could do that. If it could construct these tiny grammars, it'll still give you an infinite number of things. But, you know, [AUDIO OUT] movie names. Or things like verbing proper name turns out to be a really good thing for like, you know, Amy Stopping, Interrupting Timmy. It's so bad. And you could find that from looking at these things. And just to show you how much I think that this is, you know, actually not that bad of a problem, I did not this grammar thing. I did something even simpler, which is to take all the other names that I could find on Wikipedia for different movie genres throughout the ages, and then I looked at things like romantic dramas. And what I did was construct a very simple n-gram which just takes those words and just sort of does random walk on those words. OK? And you could imagine complicating this immediately by taking something like embedding those words in the high-dimensional space and actually picking words that aren't close to that. So you could get new words that were never in there before. I'm not even doing that. I'm doing something ridiculously simple that I don't think people are doing. But let me show you how reasonable it is. OK? And what I'm going to compare it to is some stuff that we ask people on Mechanical Turk to give us names for a new romantic drama. Ah, the only thing I forgot was the right labeling for these things. So Laura, what do you think? Is this from Turk, or is this my algorithm? LAURA SCHULZ: I am 50/50 [AUDIO OUT].. TOMER ULLMAN: So how about we'll have by, not show of hands, but people just shout it out. Like, if it's, I don't know, Turk or-- I'm looking for a short word which is like a Turk or Tomer. Let's do it that way, so Tomer just standing in for Tomer' simple silly algorithm. So who thinks that this was created by someone an actual human on Mechanical Turk? And who thinks it was created by Tomer mechanically running through an algorithm? OK. So in 3, 2, 1 you're either going to shout Turk or Tomer. 3, 2, 1. [INTERPOSING VOICES] TOMER ULLMAN: OK. That was actually someone at Mechanical Turk. Let's do this again. Girls In Ships for a romantic drama, 3, 2, 1-- AUDIENCE: Tomer. TOMER ULLMAN: This was an algorithm. Value Of Love, 3, 2, 1-- AUDIENCE: Turk. TOMER ULLMAN: That was Turk, good. Endless Love, 3, 2, 1-- AUDIENCE: Turk. TOMER ULLMAN: Good. OK. How about Legend of Paris? 3, 2, 1-- [INTERPOSING VOICES] TOMER ULLMAN: Nobody knows. This is actually me. OK. Who's enjoying this and wants to do a few more? Land of Roses, 3, 2, 1-- AUDIENCE: Tomer. TOMER ULLMAN: Tomer. And finally, Those We Meet Again, 3, 2, 1-- AUDIENCE: Turk. TOMER ULLMAN: No, it wasn't me. Oh, sorry, one more. Love Lightly, 3, 2, 1-- AUDIENCE: Turk. TOMER ULLMAN: Yeah, Turk. It seems like Turkers were actually doing better than the algorithm, which is romantic is love. And I'm just going to put something with love in the title. So who wants to do this action movies, and then we'll start stop? OK. Let's do this for action. How about The Chase? 3, 2, 1-- AUDIENCE: Turk. TOMER ULLMAN: Yes, how about Who, The Annihilation? [LAUGHTER] TOMER ULLMAN: OK, that's me. The Oversight? Turk. The Edge. AUDIENCE: Turk. TOMER ULLMAN: Turk. Jack Death? Tomer. Among Heroes. AUDIENCE: Turk. TOMER ULLMAN: No, it was me. Swordmen in China Three? AUDIENCE: Tomer TOMER ULLMAN: Tomer. And The Hit? AUDIENCE: Turk. TOMER ULLMAN: People on Turk. You can probably [AUDIO OUT] than four in each one. And, again, people are like, The Oversight, The Hit, The Chase, The Edge. That's the only thing they did. They actually came up with some clever stuff as well. But, you know, it's interesting. And, of course, I'm cheating. Because the algorithm did a bunch of really dumb stuff that I didn't put in here, like Hunchback of Monte Cristo, Get it Did, Bell of a Lesser God, Eagles Shooting Heroes, Tomb Raider, The Raging God Of Violence, and Legend of Legend, my favorite. But my point is to say, you know, in the same sense that Joshua's saying, you know, imagine that you could use something like a ConvNet to quickly cede your proposals-- imagine if you could think of like a random dumb algorithm that could then [AUDIO OUT] and say Legend of something. And then you start to say, no. That's not really a great idea. What you're trying to get at here is not 100% accuracy with these silly things, but something like 1 in 5. 1 in 5 is better than 1 in 0, or 1 in a million or something like that, which is what Laura was pointing out. OK. So as I said, we still have a long, long way to go to model children to meet Laura's critique. It's hard to say what's hard. I think that's what I was trying to hint at with Steve Piantadosi's point. We don't really know what's easy. We don't really know what's hard. But people in development and in computational land should continue to care about stochastic algorithms. And people in computational land should continue to care about children to everyone's benefit. And that's it-- so, Laura. LAURA SCHULZ: This will be very short. [APPLAUSE] So I didn't know. Or rather, I did know. But I didn't know it was going to be part of this debate about Max Siegel's thing. So I'll say something briefly about that. But you've had three really good approaches to each of these. So I'll speak briefly about them. I think what Owen is doing is totally great, but still driven in some sense by the data, not by the question, right? And I think the point that I'm going to make just continually here is that the way we think is driven by the goals that we have, right? And each of these solutions in some ways is failing to use what is most salient to it as humans, which is we have problems. We have questions, right? And what I would like to do is see us move to a case where it's not just the data that's causing us to generate new ideas. And we're not just trying to deal with that. It is actually the information in the problem itself. Similarly, I think what Eyal's doing is totally beautiful. And the representational compression is really, really interesting. But a lot of learning problems can't be solved. Most of the ones I was gesturing at are not really problems of changing the representational format. It matters hugely that we have an Arabic numeral system instead of a Roman numeral system. That changes the kinds of problems that we can solve. And so that represents a huge advancement. And for many kinds of problems, it will make search much more efficient. But a lot of problems just don't have that property. So it's, in some sense, an answer to a different kind of problem. Steve's proposal-- what can you say, right? It could be true. There are a billion, billion neurons. You get more synaptic connections than there are stars in the known universe. Of course, it could be true. That's what an expedition means-- a long line of everybody, says Pooh. But it's not as good a story if it's true. So it could be true. You could do a billion things really really fast and just think about the ones that you arrive at. But I think the jury's out on that one. Max and Tomer and this-- and while ago I think Sam Gershman also, right? So Sam Gershman came up to us and we spent a while talking about how you would invent what we were affectionately calling a bullshit generator, our ability to [AUDIO OUT]. Somebody asked me about anything, you know-- tell me about Ionic and Doric columns, you remember something from sixth grade. And you start talking, right? So the question is, what can you do? And I think this is a really nice attempt. And I think the idea of seeding it from past examples to help construct a search space is a really beautiful idea. Again, the question is, how do you make that. My feeling is still we can do something that works for those kinds of problems where we have past examples. We can do it for any kind of problem. And so what I really want to push for is use the problem, right? Use the information and the problem. Because for those problems, like romantic movies, we have some existing examples. We can the search space. We can do that. And for my theater company example, it's perfect. But for the peppermint example, not so much, right? You're not going to see the search space from examples of candies, or what you know about the construction of candies. If you did, it wouldn't generate the pendulum answer, which we think is a good wrong answer. So it's not just that. It is I have a problem of a particular kind. It is going to be satisfied by some kind of an answer and not others. How can I use that to help my [INAUDIBLE]?? So that's I think the end of what I have to say. And we'll return to questions.
MIT_RES9003_Brains_Minds_and_Machines_Summer_Course_Summer_2015	Lecture_62_Ken_Nakayama_The_Social_Mind.txt	The following content is provided under a Creative Commons License. Your support will help MIT OpenCourseWare continue to offer high quality educational resources for free. To make a donation, or view additional materials from hundreds of MIT courses, visit MITOpenCourseWare@OCW.MIT.edu. KEN NAKAYAMA: The social mind, let me see if I can just go through these slides here. One might ask the question on the origin of human intelligence. OK, I just want to give a little bit of background. Nancy summarized my career very nicely. I started out recording from retinal ganglion cells and animals. That was a very long time ago. And I was kind of into reductionism, trying to understand how retinal ganglion cells determined how we see certain psychophysical phenomenon. So what am I doing here talking about some of that social processing? Well, it's been a long time ago, so in your career you do different things. But as I'll say in a moment, I happened to read an interesting paper. I was in the University of Tokyo library kind of bored and I found a book. And I found an article, which I'll talk about that. Didn't change my life, exactly, but it reoriented me towards this social processing. And you'll see there's many different ways of studying. I think the realization that social processing and the human mind are really very inextricably connected opens the door for kinds of investigations that I think are sort of really open-ended and somewhat limited just by your imagination. So I sort of encourage some of you with that bent to think about this. Since I study vision I will just talk a little bit about the visual system as a preamble. And I was studying vision about 50 years ago. And I think I was sleeping for about 10 years. And all of a sudden I found out something interesting. Here's a picture of the visual system when I started graduate school. There was the motor system, the somatosensory system, coupled sensory system. And vision was that area 17 V1 in the cortex I think I slept a little bit and all of a sudden I woke up and oops, little bit past 1970. Almost the whole half of the brain in the back and the front was devoted to vision. You probably had lectures by that. And so I think that's really an important fact because the fact that the brain is very visual-- I mean, that's human. So maybe in mice it might be different, they may be it's old fashioned, or something like that. But for primates I think it's very important. You probably learned about the different systems. That the posterior part of the brain. There's a lot of anterior parts. David Marr said, anybody heard of David Marr? OK, David Marr said vision, where is what by looking, what is where by looking? I just want to say it's way, way more than that. The visual system-- I mean, if vision's half the brain, that's just a pretty, puny problem. So we have to think of all the things that the visual system might do. And we really haven't thought of them. Well here's some more pictures of a visual system. So, it's a little bit like, it started over here on the west and you go east. It's sort of like Russia going all the way to the Pacific Ocean. Vision system really expand. I think it's kind of slowed down recently. I would just say the visual system is very big in primates. And if it's half the brain, well all the other stuff, like greed, sex, power, music, all that other stuff. It's about the same. So, that means that we just don't really understand. That's a lot of stuff over there on the right. That means, on the left, we don't know squat. So just have that in mind, there's plenty for you to do as young scientists. So I don't know. It must be-- This is a very tiny list. It's lots of things. It's David Marr kind of thing. So it's more than that. I'm going to talk about visual motor control. I'm a psychologist. For some reason, I think a lot of our-- Roger Sperry said 70 years ago that the ultimate arbiter of your mind is action. I mean, if you don't do something-- if we can have-- I mean, I studied perception most of my life, but if you don't do something, you're not going to pass your genes on to the next generation. So action is it. Psychologists, they don't study action at all. That's like, they've forgotten that. So I just like to put a plug-in for action. There's people who study the motor system that are kind of picky. So and even if they are, and don't let you into the field, it's important field. I think you should push your way in. I think it's extremely important for navigation. Most animals live in a widely extended space. They know how to go home quickly if they're trouble and things like that. I'll talk a little bit about some kind of animals like that. The range of area that animals go over is astonishing. And so I think the understanding of animals in their natural environment, there's lots of work on it, but I think it's-- a lot of it's quite mysterious. But I'm just going to talk mostly about social perception. And this is the paper I read a long time ago. I never met this guy. His name is Nick Humphrey. And he wrote this paper, I don't know, maybe 30 years ago, and I just happened to come upon it. And it was in a book. And he said-- he was not very nice. He said, experimental psychology in Britain have tended to regard social psychology as a poor country cousin of their subject. So basically, he was putting down-- but he really wasn't because he really said something different. He said, wait a second. And he sort of turned psychology upside down. And he said, you know, actually, the social part of intelligence is the thing that drives everything else. I mean, he might be overstating it, but let's see what he says, here. Oh, I'll come back to that in a minute. The intellectual faculties of primates have evolved as an adaptation of complexity of social living. For better or worse, styles of thinking which are primarily suited to social problem- solving color the behavior of man and other primates even towards the inanimate world. So he's sort of making the implication when I said, where does intelligence come from? One of the sources of intelligence are the nature of the social world. And it's a quite a complicated world here, because this guy, Dunbar. You've heard of this guy. He sort of-- I'm not sure this is a really good study, but what he did was he made-- he got the size of the cortex of lots of different primates and he made some kind of estimate. I have no idea how he did this. How many animals were in their troop. But you can sort of imagine, if you have a different social group, it's not just-- it doesn't just go up by N. Because let's say, you know, it's not just how many people you have to deal with. Those people deal with other people and they might be plotting against you. So in other words, the number of permutations of social behavior really-- I mean, it's only one other person, it's not too complicated. As you get like six, two people might be plotting against the third one over there. All kinds of possible things happen. And as some of the primatologist, if you read some of the accounts, there's planned murders of all kinds of people in the top. It really looks a little bit like the Medici's back in the Machiavellian intelligence. Doesn't sound so weird. OK. So basically, the idea is that intelligence-- we have to think of intelligence in the social world. I'm just scratching the surface here. So let me just talk a little bit about prediction. That's kind of what we think one of the hallmarks of science is, prediction, right? So what are-- in what area is prediction better? Any thoughts. Suppose I took a rock and threw it down the mountainside. I couldn't do squat, right? I mean-- AUDIENCE: Well, you could do all the measuring, right? PROFESSOR: Yeah, but nobody's going to do that, right? AUDIENCE: That's what I'm saying. PROFESSOR: OK. Biology, I mean, we know-- for example, we know that behavior of atoms, we can't really predict the behavior of atoms over long trajectories, but we can show that, you know, at a certain date, maybe next year, at Woods Hall, people from all over the world, including Nancy, are going to show up at 9 o'clock, here. We can predict that. And we can predict things like my alarm clock is going to go off tomorrow at a certain time because I set it. So what I'm trying to talk about is kind of a point that Dan Dennett made, who is a philosopher. Who, I think, has really made really seminal contributions, at least to my understanding of psychology to a large extent because he really talks about explaining things at different levels. This is another-- I started out as a reductionist in a sense, but let me just give you, give you a sort of a thought, here. Have you ever noticed that when the moon is setting over the Western sky, it's the new moon. It sometimes seems kind of the whole moon seems to be illuminated a little bit. It's not, it's not just that crescent. If you think of the, if you think of the sun, it should only be a little crescent. So why do you see the rest of the moon filled in? Well, there's the possibility that the sun reflecting off the earth, which then sends light back to the moon. That's an explanation. Do we need to know about photons? Do we need to know about duality of waves and particles? No. That's an explanation. That's all I'm trying to say. Some explanations are pretty darn simple. And in science, I think we're looking for any fun, interesting explanation we can find. So I think Dennett has nailed it to some extent. OK. So he basically has three levels for predicting behavior. The physical stance. I mean, if we have ideal situation of Galileo dropping his balls as a flight that, assuming there's no resistance to air and stuff like that, we can make nice predictions. Biology and engineering, we know that you know, seven minutes, the coffee is going to be ready, because that's what it was designed for. But then he's got something called intentional stance, which is in order to really understand what people are going to do, Churchland would say, well, we go in to look at their nervous system. We record from their neurons and stuff like that. But Dennett would say, no, we can't really use the physics thing or even design, we just use some other heuristic, which is very explanatory, what people desire and what their beliefs are. If we know that, we really can understand what they're going to do. And of course, your grandma knew that or your great grandmother knew that as well. So but these are really great ways to understand people. And so I think that we can't just turn our backs to these things. And I think Dennett-- the question is, and we might come back to the end. What's the sort of scientific sort of status of some those types of explanations. Are they scientific? Or what are they? Maybe if we have time, we can talk about that. OK. I'll skip that part, here. So I sort of feel, just as-- what I'm saying is it's a non- reductionistic way of looking at thing-- you should open your mind. And actually, CBMM or brains, minds, and machines is kind of your kind of condition to that because some people come from computer science. I mean, you can't really understand a computer in terms of atoms and molecules. You really can only understand the computer at some kind of a higher level. So I'm just telling you, that's fine. If you're a neuroscientist and want to reduce it to synapses, things like at, good try, but it might be difficult to do. There are many other valid things to do. But one of the things to do-- let's just examine some things about them and explore them. I think, as I say, there's so many possible things in this realm. Doing anything, I mean, people have all kinds of principles. My approach in the area, this kind of area of science, or any science that I've done, is to develop new tools, and then helps you explore. OK. So there's lots of things that humans do. But what I want to stress in the first part of this talk is that our human social behavior is really not unique in many ways. There are very, there are very important things that animals do. And I'll talk about them shortly. Animals, really are very social. There are certain animals like octopus, apparently is not very intelligent, not very social, but by and large many animals are very social. Doesn't explain everything but it's-- I think we share some core things. And if we can understand some of the behavior of animals or why they're social and how their social mechanisms work, I think we have a treasure. So the area of animal behavior is very underfunded area right now. I sort of feel-- I just happen to like to watch nature videos. And I'll show you a few, you'll have to indulge me a few of these. But I sort of feel these can be more important than careful psychophysics because I think they open your mind to things that you may not have thought about. So I'm a kind of devoté of these things. OK. So do animals have an intentional stance, etc., kinds of things like that? We don't know. Here's a fun video I think maybe half of the audience have seen this video. And maybe I'll show the whole thing because I think we have time. It was taken just by chance. I think it's been a TV program now. And I think it's got like 50 million viewers. It's about different kinds of animals interacting, their social behavior. It's about animals that you do not think are that smart and things like that. Ungulates. I don't know. I don't know what buffaloes are exactly, but we've never thought of them as highly intelligent or social. AUDIENCE: That is a huge buffalo. AUDIENCE: That's a huge buffalo. AUDIENCE: That's a huge buffalo. AUDIENCE: I love this. AUDIENCE: Is it little? AUDIENCE: They're crouching. AUDIENCE: She's crouching. AUDIENCE: She's going to get eaten. AUDIENCE: Oh, my god. AUDIENCE: Oh, my god. Oh, my god. [INAUDIBLE] Oh, she's going for him, she's going for him, she's going for him. She got him. AUDIENCE: Jeez. AUDIENCE: Oh, she did. She got him. AUDIENCE: Ladies. AUDIENCE: [INAUDIBLE]. AUDIENCE: No, the lions have won. AUDIENCE: Yeah, but look at all those buffalo. AUDIENCE: --Buffalo down. AUDIENCE: They're like, go and try chase the lion, but I think they're too late. AUDIENCE: They're going to chase [INAUDIBLE].. AUDIENCE: Look at the teeth, Jay. AUDIENCE: You're too late. You're too late. AUDIENCE: I think because he cannot believe what's going here. There's a big barrier between lions, crocodiles, and buffaloes. AUDIENCE: Look at them all. AUDIENCE: Whoa. AUDIENCE: He swatted at him and kicked at him. He's kicking at him, look. He's kicking at him. AUDIENCE: [INAUDIBLE] I mean, buffaloes is basically used to [INAUDIBLE]. [INTERPOSING VOICES] AUDIENCE: And in deed, [INAUDIBLE].. AUDIENCE: Oooh, they got him surrounded. AUDIENCE: And that one's-- AUDIENCE: Ooh. AUDIENCE: Chasing-- go on, go. Chasing-- AUDIENCE: You got the lion [INAUDIBLE] right. AUDIENCE: [INAUDIBLE]. Dave, can you get the peace? AUDIENCE: The others are doing that. [INAUDIBLE] [INTERPOSING VOICES] AUDIENCE: Never seen that. AUDIENCE: I've never seen-- AUDIENCE: The calf's still alive. AUDIENCE: It is? AUDIENCE: Yeah. It's trying to get away. It's standing up. AUDIENCE: It is. It's still alive. PROFESSOR: There's much more to this. There's many other videos about lions and buffalo finding a layer of baby lions and the mother is gone. And they just kill all the lions in there for revenge, or whatever it is, your interpretation, I don't know. But anyway, so I'm just trying to tell you. I'm not going to show you more, but in the lectures, there's going to be a number of videos here. I'll show one more I think. But I think this is-- I don't know, I think you can see this is better than an experiment, OK? You really get-- to me, it really showed me a kind of, a kind of a social organization. A kind of-- are these animals conscious? All of these kinds of questions here that I think you should think about. OK. So on my website, on the lecture there's going to be some other videos that you can look at and things like that. I won't do this one, here. This is a very interesting video. Some of you might have seen it as well. I don't exactly know what the origin, I won't show it here today. But it's about a-- on the web, it says monkey teases tiger, it's a gibbon, or something like that. But it's a really amazing show. It goes on for about five minutes. We're given just swinging around and making himself ready for attack. He runs along the ground, the guys are chasing, it's like playing tag. And he runs up a tree and he jumps over them and he their tail and he grabs their ears. There are two lions, two tigers there. And he's just putting his life at risk, but he's doing it. Maybe that's that the theory is, maybe that's his territory. He doesn't want those people around there. But it's pretty amazing. So just Google monkey you know, teases tiger. You'll see the full video. It's a pretty damn amazing video. Another video that is fairly interesting and this is just-- just shows you. This is a guy named Jaak Panksepp. I used to be on a study section with this guy. I couldn't figure out what this guy's talking about. This was many years ago. And he's now an expert on laughter in rats. And he tickles rats and they follow him around. And can you hear them laugh? No, you can't hear them laugh. So he got this bat detector which takes ultrasonic sounds and brings it down to our frequency level. And you can hear the rats laughing. So try that out as well. So you can do that in your spare time. So he's got some very-- Oh. This one is kind of interesting, War. One of the interesting things about-- We have sort of a whole field called evolutionary psychology. And what's the reasons why animals might cooperate a lot? That's what we're talking about here. These are really-- a lot of these things are up in the air. And there's a lot of debates, but one possibility is, if animals cooperate, they can deal with a rival group of animals who might be taking over their territory. Animals have private property, you might call it territory. They don't have contracts or anything like that, but it's pretty close. And if they have territory that they have and there's another group that has territory, there's going to be conflicts. And why not call it war? And humans have had war for many generations. And I just like to show a little bit that-- these videos are incredible videos. The guy's name is Timothy Clutton- Brock. He's a ethologist at the University of Cambridge. He studied many different kinds of animals. And these meerkats-- there's been a, there's actually been a TV show. Have you ever watched the BBC show? Anybody here? It's a BBC show and it's gone for four years. I mean, how long did Dynasty go for? Or all these ones that you guys watch? I mean, four years of this life of these animals here. And the really interesting thing about-- they're about one foot tall and they run in troops. This group of animals is really interesting. There's more animals than you think. These are female- dominated societies. There's a alpha female. And she has, she has many children. And then her daughters are supposed to take care of her kids, not their kids. What happens is sometimes the daughter wanders off and some Romeo was coming around and they had babies and stuff like that. That daughter has to really be nice to the mom to stay. Usually she's banished and she dies, or she joins that other troop. There's so many different things. And these animals-- you can see these guys, here. See the investigators on the BBC show-- these animals somehow have been totally inured to humans. So they're doing all their thing and people are just walking around with their cameras. So this is-- you don't have to set up a laboratory, you just go out there. And you have to go to the Kalahari Desert. But I think that this is an incredible research enterprise that's been going on for about 10 years. One of the things is they-- I think, in many ways, it's one of the most studied social animals because the Kalahari Desert is just open. You can see everything. One of the things they do is they dig down into burrows. And so when they're digging way down-- they can eat scorpions. They're not, they're not bothered if they get stung by a scorpion. But if they're down there and there's hawks going over, they can just pick them off. So they have these sentries and things like that. So they sort of double up-- and then they have babysitting co-ops where they're teenagers that-- they have these burrows underground. But of course you can't-- if you're growing up as a meerkat. By the way, meerkats have nothing to do with cats, but they're somehow related to mongoose. They're sort of a little more in that category. But anyway, they have a baby sitting kind of thing where they let the kids out on there. They play a little bit and if it gets strange, they run back. And so there's a whole babysitting thing because every day they have to go out and forage for food. And they go very long distances. The territory of these meerkats could be as much as a square millimeter. That's amazing. Just think of these animals, this big, going over a square millimeter, a kilometer. That's territory. So what we have, of course, in the human situation-- DOCUMENTARY NARRATOR: Moments later, the rest of the whiskers return from foraging. From the brow of the hill, they see that their home has been overrun by the [INAUDIBLE].. The rival group spot the owners of the burrow and the angry whiskers waste no time in commencing the attack. [INTERPOSING VOICES] DOCUMENTARY NARRATOR: --Straight towards the whiskers. They're going to defend this piece of territory as if it's their own. A bloody fight is inevitable. One of these sworn enemies will have to be in [INAUDIBLE].. The [INAUDIBLE] a hasty retreat. PROFESSOR: The good guys won. What I'm going to tell you is a pale sort of version of all that stuff but I think we have to study stuff in the laboratory. So a lot of my interest comes from the work of Nalini Ambady who is a colleague of mine, long ago in psychology department. And so what-- we're going to go-- I'm going to talk about humans now because that's what I've been able to do. I'm not-- But I do think that we really need to look at all different parts of the animal kingdom. But she did something that's really quite interesting as a social psychologist. Mostly people have been handing out questionnaires and people how-- what do you think about things? Social psychology has mostly been about attitudes and had a rightful good history about prejudice. And how people who have-- a lot of the social psychology came out of the-- was stimulated by what happened in the Second World War, the atrocities and things like that. How could people do things like that? And so those are really important things. And you read about them. But she did something sort of more at an everyday level in a sense, and just take pictures, very few pictures of different people. And just see what people get from these pictures. So clip, movie clips. And basically, she did one thing and just had people rate teachers that they-- this is at Harvard. We had teaching fellows and they were just talking to their class and things like that. And then students just rated them, what they thought, how good a teacher they were, just on different adjectives. And that really predicted the student ratings of the whole course. In other words, 10 seconds was able to predict the ratings of the whole course. So remember that when you do teaching. I'm hopeless, but who knows. Anyway-- Basically, here's another thing they did. OK. There's that. Gaydar, you can tell if somebody is gay or straight or something like that. It only presenting the thing for about 15 millisecond. I find that hard to believe. If you look at a video, two people talking together, you can tell whether they're friends or strangers. There was a really interesting one. This is really-- outcome is you have an interview or tapes between doctors and patients. And you can outcome variables like whether that doctor got sued or not. You can have people rate, have adjectives and it's sort of like, a little bit like-- you can think of doing machine learning, just taking all those adjectives and then sort of predict whether that person is going to be sued. You can see just by this a kind of a crowd-sourcing way of dealing with all kinds of real world, social, significant things. OK. So I've worked a lot on face recognition. That's a very important thing in the sense that in order to negotiate our social world, if you can't do facial recognition, you're in trouble. I work a lot of people with prosopagnosic. Oliver Sacks. May know who Oliver Sacks is. He's definitely prosopagnosic. And if he's at a party, he's a little bit lost. But he's a-- he has so many other charming characteristics that he can manage. People come up and talk to him. But you can imagine, if you're not that charming, witty, funny person and you don't recognize people who you're supposed to recognize, you're in trouble. One of the subject I worked with, she's a quite a well-known author. And she goes on book tours, she's happy, and she is she does very, very well. She has a problem with face recognition, but she can't really recognize a lot of the guys in her departure. She teaches-- She's an English literature professor. There's about four guys in our department. She can't really recognize one from another. That makes social interactions very difficult. So we all are experts at face recognition but I've studied something called face blindness. But I just want to step back because the thought occurred to me. This is one thing that's really-- we've just published, not myself, but my collaborators, we've-- well, I'll come back to that later. It's called face recognition under early stress. I'll come back to that later. Remind me. OK. So over the last 100 years or so, there's something called acquired prosopagnosia which means that, I think, the best case was, I think, World War II victims, the German soldiers in the Second World War. And those people, you could tell they had face recognition problems because of course, they were able to recognize faces, and later, they weren't able to do it. And so-- but more recently, and I've studied hundreds of these people, and we've actually signed up thousands of them. People-- we just we put a website out and we basically have studied huge numbers of people who are just naturally occurring people who can't recognize faces very well at all. And we've got about 6,000 registrants. We really don't test them that much now, but we can test them on the web. As Nancy mentioned, we do that kind of thing. Just some testimonials from these people, one or two, just shows the problem they have. This week, I went to the wrong baby at my son's daycare and only realized he was not my son, blah, blah, blah. So in an audience like this, you don't have to identify yourself since you got so many people who are STEM people. STEM people sometimes have these problems. Usually if I give a lecture to an audience like that, somebody will come up and say that's me. So this is a lady I know quite well. She's got a PhD in differential geometry. She's a mathematician. She has lots of problems. She is one of the few [INAUDIBLE] that I think is kind of mildly on the spectrum. But most of the people I've studied are not autistic or Asperger's in any way. But a lot of people have problem. I claim it's a visual problem. It's not a psychiatric problem or a social problem, but it does lead to social problems. And with Brad Duchaine, I've studied these people quite extensively. And the thing that-- I just want to say, in order to study them, we developed a face recognition test. I got into the testing business in this way because the tests that were out there weren't very good. They were really bad. I mean, they were-- I won't go into how bad they were. So we spent three or four years just making up a test, which is now used widely around the world. And I don't want to go into the details of the test but it basically enables people to sort of over-learn about six faces. It's a little bit like natural face recognition. It's not-- I think it's a very good test. I won't go into why and things like that. And it gets harder and harder. But I just want to say that we've been able to characterize these prosopagnosia people. But somebody all of a sudden shows up and says, well, I'm not one of them. I'm super. So we've study those people as well. I won't go into detail. We made a lot of studies about them. Just a couple of testimonials. I pretend I don't remember people because people think I'm stalking them, or something like that. So now we've studied half a dozen of them. And now we're studying a much larger population of them. I would-- All I'd like to say is these people are as good as prosopagnosics are bad. They're really-- but we haven't found that many of them. We've probably identified about 30 of them and we've been looking. And we give them much harder tests of face recognition. Some of you, if you can recognize all four of these people before they were famous, I'll come up and you might be one of our people, but I doubt it. Most people think they're very good at face recognition. When we test them, they're not that good. So nature versus nurture. Make sure I don't go over my time, here. You know, most of you are not psychol-- how many people are psychologists here? OK. So you know that there's some experiments called twin studies where you study some characteristics in monozygotic twins and then different, dizygotic, dizygotic twins. And so the idea here is that they sort of share the same family environment. Of course, each person's different, and they have a different way of dealing with the family and stuff like that. But that's the best we can do. But if you can show that the correlation between twin 1 and twin 2 is very, very high, and in our case, it's going to be extremely high. In fact, it's just as high if twin 1 took the test twice. That indicates-- and then if you compare it with dizygotic twins, there's a real difference there. And that's what we've shown essentially, is that-- I'm just saying, the test reliability is very good. It's tech has a R of 0.7, which is very good. And we had-- There's 350 just random people who took the test twice. But now what you can do, we have this online way of doing research where we can just have people come up and-- the real hard thing in twin studies is not to get monozygotic twins, it's to get dizygotic twins. Dizygotic twins don't really care about being twins that much, but monozygotic feel, hey, I'm a twin. My wife is a monozygotic twin so I know all about this. So we have something called the Austrian twin registry. And we pay them and they-- a very nominal fee and they do our tests online. And what's really interesting, we found if you correlate twin 1 and twin 2, they have to be obviously the same gender. The correlation, you'll see here, is 0.7. It's just as good if the twin took the test twice. That's pretty darn amazing. And the dizygotic is 0.3. So in this particular situation, I think we've shown to our astonishment that the ability of this face-- the ability to learn new faces is almost entirely heritable, whatever that means. I mean, it's a technical definition, but it indicates that your ability to recognize faces is strongly controlled by other factors than experience. And that's when I want to bring up this childhood adversity thing, here. Laura Germine has done a very large survey of I think, over 1,000 people who were-- and these are things-- we had trouble with the IRB here, but people talked about all of their horrible things that happened to them, childhood sexual abuse, neglect by their parents, all kinds of things that are really bad when you're growing up. And then we did a lot of tests on the web. And everything, all our tests showed real deficits in cognitive abilities, in even emotion recognition, but there was no deficit in ability to learn faces. So that's pretty-- so what I would like to suggest is that somehow we're sort of topped out. We have enough experience seeing faces and we've reached our asymptote. That's just a point that might be of interest here. So another thing we've done recently, I don't have any data here, but we just got it paper accepted, was to show that using the same sort of paradigm of monozygotic and dizygotic twins that face attractiveness is not-- there's no genetic component at all, basically zero. So here's a face attractiveness test that we've cooked up here. Here's a bunch of faces. You have a bunch of cards, you sort them. That's all you do. You get the mean ratings for all the faces and then you rate the faces. And I'm going to say, OK, you're-- let's say that upper left- hand corner, that's one person's ratings versus the mean ratings. That's a correlation of 0.74. The next person's 0.91. That's really unimaginable to me that, that person there's correlation with the mean rating is so high. And look at these ratings. They're very, very high. Of course, these are Harvard undergraduates. They all have the same kind of mentality. Perhaps that's maybe that's the point. But the correlation with the mean, this is just from one of my courses. It's unbelievable. The correlation, the agreement of who's attractive and who's not attractive is astonishing. So what we did was we had these MZ versus DZ twins and had them rate them. We have a kind of a jingle. We say, how quirky are you when you're a beauty judge, or something like that because our web site, we don't pay anybody. In the twin study, we do, but we like to make the tests fun and interesting to the people. And we, most of the time, give them feedback as to how they done. So all I'm trying to tell you there is that we don't see any genetic component there at all when you do the monozygotic and dizygotic twins, here. Gender versus attractiveness. Where you give a bunch of pictures to people and you say, which looks more feminine, more masculine? You do that with girls and then you do it with boys. Which was more masculine, feminine? Those are very reliable ratings also. It's hard to see but female attractiveness correlates very well with the gender judgment, male attractiveness doesn't. Actually male attractiveness is quite robust. People can agree on male attractiveness, but it's not the gender axis. Maybe you know, 2000 years ago it was machoness or something like that. But in our society, it's agreed upon but it's a little bit mysterious. Alex Todorov has published a paper on this, which I don't understand. But he says he explains it, but I'm not sure. But that's an interesting area of what constitutes male attractiveness. A fellow, who is an MIT student, Richard Russell, found one of the things that is related to female attractiveness. It's basically contrast, facial contrast. Not how dark you are, but your facial contrast. And he just found that if you just measured the contrast of pictures, you find the female average contrast is higher than the male contrast. And if you take a picture like that, which is an androgynous person here, and you just change-- you have the same picture and you just increase the contrast, the one on the left looks more female. And maybe that's the reason that Richard thinks that this is the basis of some cosmetics. And he works with a cosmetic company. He's not doing-- OK. So the final part of this talk, I'd like to talk about is that, which I'm more involved in right now. Is I sort of feel that most of social psychology is involved in kind of asking people questions and things like that about sociology. But as I said in the beginning of the talk, I feel that the realm of human action is something that has not been investigated. So I just feel it's something that we need to probe. And what I would like to argue is that by studying our human actions, we can actually reveal our social perceptions. That's kind of the point here. So most of you have been in very crowded places and you don't bump into people and stuff like that. And there's many reasons for that. But basically it's pretty amazing that people, pretty much, don't run into each other in very crowded and when they're very rushed. And there are many situations in which your social interactions with people are extremely skilled in a sense, in different kinds of dancing or fencing or different kinds of athletics. So I claim this is the domain of what I call rapid social perception. So how can we study rapid social perception? Well, I'm OK with all these things, showing movies and stuff like that. But we've sort of put it into the laboratory, in a more discrete fashion just because we can study it easily. Now one of them is that most of you watch the World Cup, is the penalty kick. It doesn't happen all the time, but every once in a while, it happens. And you know, that's a really dramatic moment. And of course, you have to predict which way the kicker is going to go ahead of time if you're the goalie. So we've set up something which was sort of parallel to it which I call the lab version, here. So here's the kicker, here. But the job is not to kick the ball into the area, but just touch target left or right. That's your job. That's all you do. Go boom, boom. That's it. That's the trial. So it's very simple. And we tell people it's a game. If you are up to the game-- The blocker-- if you get there within 150 mil-- we sort of titrate it to match the skill of the two people. But it's a game. Harvard undergraduates love the game. They get tired because a little-- but there's no problem. Harvard undergraduates don't know each other usually so they're not friends or anything like that. They just, they just come in there. So I think there's a movie here as well. Let's see if I can get this going here. Oh, here, go this way. OK. So this is the game, here. Is it moving? There you go. Here. That's it. That's the game. I like pretty simple experiments. OK. So but then what we do is we just measure the finger position of the two players. You can just buy these things. They are off the shelf. They're not very expensive. The one I had like you know, a couple of thousand dollars these days. This is very low tech. It measures the position of the fingers and at a high precision, very fast, and stuff like that. You can buy this thing. It takes a while to get it going and stuff like that. So basically, we can map the trajectories of the various people. And you can see that they're well- behaved. And just like that. So what's interesting is the timing here. So the kicker goes like this and the blocker-- so you can make little measurements, here. And so we can just measure the horizontal positions, here. So the blue is the kicker and the red is the blocker. And so obviously, they can't do at the same time. But the reaction times are interesting. What's that difference in the launch point, there. And what we found was it was actually very low. You can see it there. Some were as low as 100 milliseconds but it's 150-- it's lower than choice reaction times normally. So if you take-- if you could have a button box and put a buzzer or something like that, choice reaction times are usually about 100 milliseconds longer. So what's going on here? But maybe choice reaction time isn't a really good choice. So what we did was we took-- we actually can take the position of the finger and map it on the screen. And we have a little experiment here, where you actually-- instead of playing against a person, you just, there's a little dot that zooms up there and you do that. So maybe it's that-- the fact of that. So we do this, and that's the experiment. And it turns out this experiment, we're about 100 milliseconds longer. So in other words, if you're playing against the human being, that's on average about 100 milliseconds. So you are faster when you're dealing with human being. I will just skip over some of this stuff, here. And again, there's no learning here. Basically, it's just, it's flat. So what's going on here? We think what's happening here, and this is the punch line, I'll just elaborate a little bit more. You think all through your life you've seen people do different kinds of actions. And if you do any kind of action, there's all kinds of postural adjustments. I was just talking to one of the guys in Emilio Bizzi's lab. If you're sort of in different kinds of sports, if you lunge like that, what's the first muscles that contract? Anybody want to know? Your butt. Because that's to maintain your center of gravity. So there's all kinds of-- I'm not saying people will watch you know, their butts and stuff like that. But when you make any kind of motion, it's your whole body is connected. And somehow we have knowledge, implicitly of this. So what's going on here? We talk about 90 milliseconds or so. What section of the body informative? Well, in a sense this is kind of-- you just block off different pieces and stuff like that. And we just, we have all different kinds of ways we can just-- limited. And we do-- we did all kinds of experiments like that. And basically, if you show all, there's an advantage, just the torso or just the head and there's the computer, it's all over the place. That the information is all over. But that's not unreasonable because when you're doing something, all parts of your body are connected. So there's quite an advantage. I think basically, oh, this is your question. We're moving this stuff. So we have-- we call them cut videos here. So what we do is we play a video, here. By the way, we now have shown that people would behave exactly the same to a video than they do to the other person. So that makes the life a little simpler. So there we go. There's-- I don't know if you can tell, but essentially, nobody, except a couple of star athletes, could tell what it was. And it turns out, in the cut videos, in other words, you don't have that preparatory stuff. You just have all that if the person is still, we don't even know what those motions are. But basically, they're 100 milliseconds slower than the cut videos. So another thing is, oh, we even did stuff-- we did another series of experiments. Well, maybe it's eye movements. And basically, we just did some more blocking. And shoulders are important, you can see, here. Oh, I didn't show that. I don't have the data there. But basically every-- even the head, there's a little-- all of those, the first five, except the last one, were very good. Even the last one, you can hardly see anything. You just see the head then you see the eye movement is a little bit of an advantage, even there. OK. So basically, what I'm trying to say more broadly, is that our ability to understand humans probably more to predict what they're going to do is something that we've learned somewhat unconsciously. We're going to do a lot of machine learning of what parts of the body it is. And all kinds of stuff like that. I'm interested in if you people have some ideas of what other kinds of things we can do to generalize this. But I think it's an area where we-- it helps us understand all the kind of knowledge that we do have of other people that we can weed out in ways that are very subtle, but reliable.
Nursing_Skills_Videos	Blood_Pressure_Measurement_How_to_Check_Blood_Pressure_Manually.txt	hey everyone it's sarah thread sterner sori and calm and today I want to demonstrate how to check a blood pressure manually first you'll need to perform hand hygiene and gather your supplies you'll need a stethoscope and a manual blood pressure cuff so let's measure a blood pressure manually to do that we want to make sure a patient is sitting down with their arm at heart level and their legs are uncross now they're lying in bed you would want to make sure that this arm is at heart level then what we're going to do is we are going to get our stethoscope and our blood pressure cuff and you want to make sure you get the right size cuff for your patients arm because if you use too big of a cuff or too small of a company can throw off the reading and what we're gonna do is we're gonna palpate the brachial artery because this is the artery we're gonna be listening to to get our blood pressure because we're going to be getting our systolic number which is that top number and this is the first sound we hear and then our diastolic number which is the bottom number and this is the point where we no longer hear the sound so whenever we're looking at the gauge of our blood pressure cuff we want to make sure we're really noting those points because it's gonna tell us our systolic and diastolic number so what we're going to do is we're going to put our cuff on our patient and we want to make sure we find the brachial artery this is the artery we palpate that we'll be using to determine our blood pressure and it's found in the bend of the arm so we're going to find it and it is located here and we're going to look on our cuff and our cuff has these arrows and because this is the left arm we're going to make sure that this arrow is pointing in that direction of where that artery is so you're gonna put the cuff up about two inches above the bend of the arm first what we want to do is we want to estimate the systolic pressure so we want to find that number to do that we're going to palpate the brachial artery and we're going to inflate the cuff until I no longer feel the brachial artery and point when I no longer feel it I need to make sure I'm looking at this gauge to know that number because that number is our estimated systolic pressure number then when I go to take the blood pressure I'm going to inflate the cuff 30 millimeters of mercury more than the estimated number now the whole reason for doing that is because we want to avoid missing the auscultate gap that can occur in some patients on all patients have it but some it's usually patients with hypertension because the oskol Tauri gap is like this abnormal silence that can occur and it will throw off whenever you actually hear that first sound which is your systolic number so I'm inflating the cuff by filling on the artery and I'm going to note the point where I no longer feel the artery which is about at the hundred then I'm going to deflate it completely and wait about thirty to sixty seconds and then we'll take the blood pressure so we're estimated systolic number is a hundred now I'm going to inflate the cuff to a hundred and thirty and that will avoid missing the oscillatory gap if one was present so I'm going to take my stethoscope put it in my ears you can use the bell or the diaphragm of your stethoscope I like to use the Bell because it's best at picking up low pitched noises so we're going to place that over the brachial artery do it lightly don't fully compress it because you can include the artery then we're going to inflate our cuff to a hundred and thirty millimeters of mercury and we're going to let it fall about two millimeters of mercury per second and we're listening for that first sandwiches our systolic number okay is 104 and we're listening for that last sound and it was 78 so the blood pressure is 104 over 78 then once you have your reading make sure you fully deflate the cuff full of air and you're going to take the cuff off of your patient of course and clean it if it's not disposable and you will document the blood pressure and what arm you took it in now water normal blood pressure readings according to the American College of Cardiology 2017 updated guidelines a normal blood pressure is a systolic less than 120 and a diastolic less than 80 elevated blood pressure would be considered a systolic of 120 to 129 and a diastolic less than 80 hypertension stage one would be a systolic of 130 to 139 or a diastolic 80 to 89 and hypertension stage 2 would be a systolic greater than 140 and a diastolic greater than 90 okay so that is how to check a blood pressure manually thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	Six_Cardinal_Fields_of_Gaze_Nursing_Nystagmus_Eyes_Cranial_Nerve_346_Test.txt	this is cereth registered nurse re and calm and in this video I'm going to demonstrate how to assess the six cardinal fields of gays now why do we assess the six Cardinal fields of gays well it tells us how well those extra ocular muscles of the eyes are working and there's six of them along with allowing us to assess cranial nerves three four and six so what I muscles are we assessing we're assessing the superior rectus muscle which helps us with upward movement of the eye the superior oblique which helps with downward and outward movement of the eye lateral rectus which helps with outward movement medial rectus which helps with inward movement inferior oblique which helps with upward and outward movement and lastly the inferior rectus which helps with downward movement and when we're performing this skill we're going to have the patient move their eyes in the following directions and the right upper part the left lower the left upper the right lower and then left to right and while the patient is doing this we are looking for a smooth motion of the eye which would be a normal finding we would not want the eyes to jerk or to shake as they move and if they did that would be known as nystagmus and how we're gonna do that was we're gonna take our pin line we're gonna hold it about twelve to fourteen inches away from the patient's nose and then what I want you to do is keep your head still don't move your head and just use your eyes to watch where I move the pin line and as you're doing this you're going to do you're going to perform it in the six cardinal fields of gaze and you're just going to move it and you're looking for any involuntary shaking of the eyes so here we go okay so that is how you assess the six cardinal fields of gays thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	12_Lead_ECG_Placement_of_Electrodes_EKG_Sticker_Lead_Procedure.txt	hey everyone it's Sarah threats Turner sorry and calm and in this video I want to demonstrate for you the lead placement for a 12-lead EKG first you want to gather your supplies so you don't want to get 10 of these electrode stickers next you'll want to get some alcohol prep this will help you remove those oils from that top layer of the skin so the electrode stickers will stick then you'll want to get a strip of abrasive tape which is just really this little fine gritty sand paper stuff that you'll use to rough up that top layer of the epidermis so these electrodes will stick to now if you're doing this on a patient who has a lot of hair on their chest you'll want to get some trimmers to trim that hair because sticking these electrode stickers over hair does not work and it causes a lot of artifacts so you'll want to trim those hairs okay so the first thing what we're gonna do is we're gonna place our limb leads which are which include our right arm or left arm or right leg and our left leg and then we will place our limb leads which will be v2 through v6 okay whenever you're placing limb leads you can either put them on the chest which you're gonna see me doing this video or you can put them on the extremities from where I worked in a stress lab we always use the chest for the limb leads because the patient was going to be exercising and when you exercise you move your arms and legs and that can cause a lot of artifact so to do the right arm we're gonna go right below the right clavicle and what we want to do is we want to prep that site you're going to take some alcohol prep cleanse it very well removing any oils from the skin let that dry then take your brace of tape and just gently rough up the skin that top layer so the sticker will sit good to it the electrode and each electrode has backing so you're gonna peel that backing off like that has like this gel and make sure they're not dry if they're really dry looking get new electrodes because it'll affect your connection and you're just gonna put that there and just smooth it on the skin now we're gonna put our left arm and you're going to do the same prep cleanse the skin with alcohol prep rough it up we're on Rob below that left clavicle and then we're gonna put our electrode okay now let's do our leg leads okay we're gonna do the right leg so we're going to go down to the right upper quadrant and again you could place this on the leg I'm going to cleanse the area I'm going to gently rough up the skin and put our electrode and we're going in the right upper quadrant now we're gonna do the last leg we're gonna go in the left upper quadrant again cleanse the area rough up the skin and place our electrode now let's place our test leads which again is v1 through v6 and to do that we need to find our intercostal spaces and it's best to follow this way of identifying the landmarks to find why intercostal space you're in so first what we're gonna do is we're gonna find the sternal notch which is literally is a notch and it's found in between the clavicles and then you're gonna go down just slightly and you're gonna feel this protrusion a little hump this is called the angle of Lewis also called the sternal angle and we're gonna place v1 first so we're gonna go to the right of that angle and we're going to be in the second intercostal space now v1 is in the fourth intercostal space right of the sternum so we're gonna go down there's a third there's the fourth so we're in the fourth intercostal space and this is where v1 is gonna go slightly just next to the sternum so again just gonna cleanse the area really good rough it up and then place your electrode okay now v2 is going to be literally right beside of it on the opposite side it's going to be again buying your sternal notch go down where the hump is which is your angle of Luis from the second intercostal space third and then fourth and it's going to be on the left side so cleanse the area and rough it up then place your electrode okay so right now we have v1 and v2 okay we're gonna skip v3 for a second you're gonna see why we're gonna go to v4 v 4 is found at that fifth intercostal space so again find your landmarks second third fourth fifth and we're gonna go mid clavicular ly so mid way of the clavicle and it's going to be right here so that is going to be v4 so we're gonna cleanse the skin rough it up and then we're going to place our electrode okay and now we're going to go back to V 3 and V 3 is in between v2 and v4 so we're going to go right there so we're gonna cleanse the skin wrap it up place our electrode okay so we have v1 v2 v3 v4 now we're going to do these five and these five is going to be level with v4 but it's going to be at the left anterior mid-axillary line so right here so cleanse area fit up man we're going to place our electrode okay and our last one is v6 and these six is going to be level with v5 but it's gonna be at the mid-axillary line so literally right underneath the armpit so it's going to be right here so we're going to cleanse it up our electrode okay so let's go over them again we have our limb leads right arm left arm right leg left leg now we have our test leads we have the 1 v2 v3 v4 v5 and v6 ok so that is the lead placement for a 12 lead EKG thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	Head_and_Neck_Assessment_Nursing_Head_to_Toe_Assessment_of_Head_Neck_ENT_Lymphatic_Cranial_Nerves.txt	this is cereth registered nurse ari ENCOM and in this video i'm going to demonstrate how to perform a head and neck assessment now if you would like to see a complete head-to-toe assessment you can check out this card up here in the corner or in the youtube description below to access the video on how to do that now before you do a head and neck assessment you'll want to provide privacy wash your hands and explain to the patient what you'll be doing and you'll want to gather your supplies you'll need an otoscope to look at the ears a tongue blade to look in the mouth a pin light to assess the eyes and gloves so let's get started we are first going to inspect the head and we are looking at the skin color he it's nice and pink we're also going to make sure that the head is the same size as how it should be for the body and it is and we're looking for any abnormal movements or twitching of the face that he can't control that are involuntary and we don't see anything and we're making sure that the face is symmetrical there's no drooping on one side like in this picture there's drooping on one side of the face and this can be seen in Bell's palsy or in stroke and we're also just looking at the eyes in the ears or they have the same level and while we're here we're going to go ahead and look at the facial expressions and test cranial nerve seven which is the facial nerve so can you close your eyes tightly for me and open them up okay now smile for me frown and puff out your cheeks okay and he did that with E so that cranial nerve is intact next what we're gonna do is we're gonna palpate the head the cranium we're gonna check for any masses indentations look for skin breakdown any infestations and for this part I like to wear gloves so let's look at the hair so what we're doing is we're filling for any masses indentations and also with this we're looking for any skin and breakdown and if your patients and mobile you really want to check the back of the head back here because they're laying on it a law and there can be break down back there also while you're doing that look inside the hair make sure there is no infestations like lice and there's no abrupt like rounding areas of baldness which could represent alopecia then after that since this patient has a beard you want to check the beard as well any lesions any infestations or anything like that and just look around and then once you're done with that what you want to do is you'll doff your gloves and perform hand hygiene next what we're going to do is we're going to find the temporal artery and we're going to palpate them bilaterally and they are both found right here and that his are about a 2 plus and then while we're right there we're gonna go ahead and test cranial nerve 5 which is the trigeminal nerve and this nerve is responsible for many things like massification so what I want to have you do B in is I want to have you clench your teeth like bite down for me and I'm going to feel the Meseta muscle which is right there this should be a nice firm ball and then feel the temporal muscle now what I'm going to do to also test that nervous have him try to open his mouth against resistance so try to do that for me okay and he can do that now while we're here we're going to go ahead and feel the temporal mandibular joint and we're gonna feel right here I'm gonna have you open and close your mouth and I'm feeling for any grading or clicking sensations and I feel none then we're gonna palpate the sinuses and I'm going to put pressure on these two sinuses right here and you tell me if you feel any pain okay so the max max look maxillary yeah in the frontal though next we're moving down to the eyes and we're going to inspect the eyes first and we're looking at several things we're looking at the eyelid we're looking at the sclera which is the white of the eyes we're looking at the iris we're looking at the pupil and we're looking at the conjunctiva so you shouldn't see any swelling of the eyelids you should see that the sclera is Y and shiny it shouldn't be yellow like in jaundice and the conjunctiva when you pull down the lower lid have don't look up it should be nice in pink it shouldn't be red you shouldn't see any drainage or anything like that in look at the eyes how do they set in the eye socket is are they equal four instances is there any strabismus is there a cross eye where one eye turns in more turns out or up or down and these eyes are normal there's no strabismus next you want to look at anisocoria where you have where one pupil would be smaller than the other people are they equal in size normal pupils should be 3 to 5 millimeters in their measurement and here his are about a 3 and they are equal next what we're gonna do is we're going to assess some cranial nerves we're gonna be looking at cranial nerve three which is ocular motor for troq Euler and then 6 which is abducens and we're gonna do several tests to check their function the first one what we're going to do is we're going to be looking for any involuntary shaking of the eye called nystagmus and how we're gonna do that was we're gonna take our pin light we're gonna hold it about 12 to 14 inches away from the patient's nose and then what I want you to do is keep your head still don't move your head and just use your eyes to watch where I move the pin line and as you're doing this you're going to do you're going to perform it in the six cardinal fields of gaze and you're just going to move it you're looking for any involuntary shaking of the eyes so here we go next we're going to see how reactive the pupils are to lie and to do that we're gonna dim the lights a little bit and we're gonna have the patient's stare off at a distant object that helps dilate those pupils and then we're going to shine using our pin light in at the side and we're gonna see how that pupil response is she constrict and then on the other side it should constrict as well so say their baseline pupil size was like three millimeters it should go down to one milliliter and it should happen on both sides okay so being stare off at that object rod on the wall over there for me okay and that dilates the pupils and we're just going to shine light in at this side okay constrict constrict okay I'm dilate again then go over to the other side do the same again and they both constricted and equal size next what we're gonna do is we're gonna check for accomodation and how we do that is we turn the lights back on we just previously had them dim but we now make it light again we're gonna have him stare off at a distant object that helps dilate the pupils and we're gonna take a pin like you can use a pin light finger and you're just gonna slowly move it inward to the nose and what you're looking for is that those pupils constrict they accommodate and the eyes cross while looking at the pin line so here we go stare off in the distance please and I don't want you to move your head or anything just keep it real still and just follow this pin light okay ready okay so now we can document because we just checked all of the things with the eyes we can document that the pupils are equal round reactive to lie in the accommodate so that's where that acronym perrla comes into play next we're going to move on to the ears so first what we do is we inspect the ears we look on the outside of the ear is there any abnormalities any redness any drainage anything like that and then are you having any pain in your ear okay and sometimes if you have patients who've had long-term gout on the helix of the ear they may have what's called a Toph I which is an accumulation of like a whitish yellowish uric acid crystals on the skin so if you ever see that that is what that looks like next we're going to palpate on the ear we're just gonna move it around and then tell me if you have any tenderness whenever I do that and any feel any abnormal masses or lesions and then move the targets a little bit does that hurt or anything like that okay so no pain or tenderness then we're going to palpate the mastoid process which is the big behind the ear and we're looking at it is it swollen is there any redness and whenever I touch on it again does it hurt okay and just see if the patient reports any tenderness with that then when you're there you can use the otoscope to inspect the tympanic membrane and remember the tympanic membrane should be a pearly gray translucent color and should be shiny so for an adult you're going to pull the pin of the ear up and back and we're just going to inspect it and also while we're looking at that we are looking at the cone of light and remember the cone of lie and the right ear should be at five o'clock and in the left ear should be at seven o'clock next we're gonna do one more thing with the ear we're gonna test cranial nerve eight which is the best tibula cochlear nerve and what I'm going to do is I'm going to include one of his ears and then whisper two words on the other side he needs to tell me what I said so you ready okay I'm gonna clued this one okay very good okay and that nerve is intact next we're gonna move on to the nose and we're going to inspect the nose we're gonna make sure its midline on the face which it is we're gonna look at the septum is it deviated anything like that and ask the patient or you have any trouble with your nose are you having any drainage or anything like that no and you want them to make you want to check the patency of the nose I mean I'm gonna have you include one side of the nostril breathe out of the other and vice versa okay heard airflow airflow nice and Paton because sometimes people can have polyps that can block it or the deviated septum then you want to take your pin light and you just want to look inside the nose look for any drainage redness or any like polyps or anything like that and everything looks clear don't see anything and then we're going to test the olfactory cranial nerve one the sense of smell so Ben what I'm gonna have you do is I'm gonna have you close your eyes and I'm gonna put something in front of your nose and have you breathe in and smell and you tell me what you smell and whenever you do this use something that's pleasant smelling not something that's really stinky cuz it could elicit like a gag reflex or something like that if the person has a sensitive nose okay okay and this was vanilla extract and that's correct so that cranial nerve is intact next we're gonna move on to the mouth and for this part like to wear gloves and if your patient is coughing and hacking you might want to wear a mask with a shield so you don't get any mucus on your face or in your mucous membranes so first what we're gonna do we're just inspecting the lips make sure they're a nice pink color they're not chapped there's no sores on them and when thing with a lot of patience whenever there oxygen saturations our load their lips may turn dusky or a blue color so you want to make sure they're nice and pink cuz that can represent our oxygen level now let's inspect the inside of the mouth but first let's test cranial nerve 12 which is the hypoglossal nerve and what I'm gonna have you do being is I'm gonna have you stick out your tongue and move it side to side okay and he does that with ease now what we're gonna do is we're going to inspect the inside the mouth you'll need a tongue blade for that and just open up your mouth for me and I'm gonna look on the inside of the cheeks nice and pink don't see any sores you're looking to see if they're nice and pink and there's no lesions or anything like that and stick out your tongue for me the tongue should be moist like this and pink you don't want to be beefy red which is like an pernicious anemia you don't want to be dry or crack that can be dehydration okay you can put the tongue yep then I want you to lift up your tongue for me and look for any lesions underneath the time that's where mouth cancer can hang out and I don't see any okay you can close then you'll why you're also looking at the gums open up a little bit you're gonna look around for cavities any loose or broken teeth no dental caries in there then okay so to open up your mouth a little bit more put your tongue down and you're gonna look at the soft and hard palate now while you're in there you want to look at the uvula make sure it is nice in midline and his is nice and midline and we're going to test cranial nerve nine the glaucio pharyngeal and so what I'm gonna do is I'm gonna have you say ah and what you want is that you've left to move up okay and then we're just gonna test the gag reflex I'm sort of just gonna poke a little bit back there and elicit a gag reply okay there you go cats really good and cranial nerve 10 the vagus is intact because he's able to talk with some talk to me without hoarseness and he's able to swallow then when you're done inspecting the mouth be sure you take off your gloves and perform hand hygiene now moving on to the neck so what we're going to do is we're going to inspect the neck for so you're gonna have the patient extend the neck up a little bit and you're looking at that trachea is it midline look for any lesions and look for any lumps like what you might see and thyroid problems like AG order and we don't see any of that then what we're gonna do is we're gonna test cranial nerve 11 which is the accessory nerve so being what I'm gonna have you do is move your head side to side up and down okay and then shrug try to shrug against my resistance and he does that with ease so that nerve is intact then we're gonna place him at a 45 degree angle and we're gonna have him turn his head to the side and what we're looking at is the jugular vein we're looking for any jugular vein distension jvd so Ben I'm gonna just turn your head to the side like that and we're looking for any distinction of the vein and we do not see any next what we're gonna do is we're going to palpate so we're gonna palpate that trachea just to confirm it is midline and Ben do you feel any tenderness or anything like that ask him if he feels any tenderness and I don't feel any lumps then next what we're gonna do is we're going to palpate the lymph nodes all sites of those and being as I do this tell me if you feel any tenderness and what I'm feeling for is any hard lumps or anything that may be inflamed so what we're gonna do turn a little bit this way and there we go we're gonna start at the pre auricular which is right in front of the ears then we're going to go to the back of the ears the posts are regular then we're going to go to the opposite battle the parotid jugular digastric then we're going to go to the submandibular and then the submental then we're gonna go to the superficial cervical and then we're going to make our way down to the deep cervical chain and you tenderness so far then we're gonna go to the posterior cervical and then right above the clavicle we're gonna go to the supraclavicular and did not feel anything and no tenderness next we're gonna palpate the carotid artery and this is one artery that you do not palpate bilaterally you do one individually so we're gonna fill on this side and you're gonna find it next to where the groove of the neck and next to the trachea and his is nice and bounding it's two plus then we're just going to fill on the other side and same strength two plus then lastly what we want to do is we're going to auscultate the carotid artery and you're going to do one side at a time and you're going to compare sides and you're gonna listen with the Bell of your stethoscope and we're listening for a bruit which is a swooshing sound so being what I want to do is I'm going to have you breathe in breathe out and hold it for me okay go breathe in breathe out okay breathe normally now did not hear it on that side okay breathe in breathe out for me and hold it okay and I did not hear a bruit on that side as well so that wraps up how to assess the head and neck and don't forget to check out that video that demonstrates the complete head-to-toe assessment thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	2Point_Gait_Crutches_Walking_Pattern_Demonstration_Nursing_Skill_NCLEX.txt	hey everyone it's sarah thread sterner sorry and calm and today we're going to demonstrate how to do the two-point gate using crutches so the two point we're gonna have two points on the ground at a time whether it's a crutch or a foot so what does it look like well this is where the patient will move the crutch on the injured side so we're going to say it's the right side so they move the right crutch and they move the left foot together then they will move the left crutch on that non injured side and the right foot together so you have two points thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	Surgical_Staple_Removal_Nursing_How_to_Remove_Surgical_Staples.txt	this is cereth registered nurse Orion doc home and in this video I want to demonstrate how to remove surgical staples so before you do this procedure of course you have to verify the physician's order because you will get an order to remove the surgical staples and pay attention to what that order says because some physicians just want you to remove every other staple while some physicians want you to remove all of them and if you do remove just every other staple the patient whenever they're discharged and have their first post-op visit they will have rest of the staples removed the next you'll want to talk to the patient explain to them what you're going to be doing and get their consent to remove the surgical staples and let them know that this procedure typically is painless they may feel like tugging or pulling at the surgical site you can offer them pain medication if it's ordered because you know this is a wound that can be tender and then you're going to gather your supplies now always check with your Hospital protocols on how they want you to remove surgical staples because some hospitals require that you wear sterile gloves when you do this others say you can wear clean gloves so follow that here we're going to be doing it with sterile gloves so it's best to get a dressing change tray which includes a lot of the stuff you need and include your sterile gloves your drape for your supplies you're in a septic to clean the surgical site before and after and extra gauze and things like that another thing you'll need is steri-strips and why series strips well one complication that can happen whenever you're removing surgical staples or even sutures is that the wound can open up called wound dehiscence and if that does happen when you do remove a staple or a suture you want to stop cover it up with a sterile dressing and notify the physician immediately but the steri-strips will help prevent that and just maintain wound integrity and we'll be applying those after we've removed the staples and of course you will need a skin staple remover kit and in this kit will come with the staple remover also has alcohol prep and Gaulle's and then when you're ready to start removing the state what you want to do first of course is perform hand hygiene and if the patient has a dressing over the surgical site where you're going to be removing the staples you can dawn clean gloves remove that old dressing then perform hand hygiene again and get ready to open up your sterile kit to remove the staples now after you do that you want to take a good look at the wound look at it does it look infected is it extremely red like this which can indicate that there's infection if so you need to let the physician know or does the wound look like it's not really ready for the staples to come out you can just take a look at it and notice if you remove that staple that wound may open up again let the position know and an important thing to remember is that staples can be in all variety of directions like friends and look at these pictures here you can have them up and down the abdomen you can have them in a sense crossing they can be in all different directions so always be aware of that as well so first what we're gonna do is we're going to get everything prepared and prepped so we're going to open up our sterile dressing tray you got to be real careful not to break sterile field with this so we're gonna open it up and on the top we have our gloves which we can take out hit right here we have our drape which we're going to carefully grab and lay right here because we're going to spread that out and you also have your alcohol prep and some other supplies so what we're going to do now is open up our drape so we're just going to gently open it and we have about two inches on the inside to be able to reach so we can open it up then what we're gonna do is we're gonna open up our supplies over here in the packages and gently drop it on to the sterile field because we'll need to be using those and we want them to stay sterile as well so we're going to open our skin stapler and remover and just gently drop it into the field discard that we're gonna open up our steri-strips as well and then discard that now we're going to Don our sterile gloves and once we have these long we want to be careful just to really touch the things that are in our sterile field because if we touch something on the outside it will contaminate our gloves if you're not familiar with how to put on sterile gloves please check out my other video on how to do that so we're gonna open those up and we're just gonna pull the tabs because we don't want to touch the gloves we can touch the inside of the gloves the cup for not the outside I'm going to glove my right hand first I'm gonna grab this part of the cuff and just slide it over a little bit then I want to take this other hand slide it underneath the other glove and glove my left hand and a lot of times you have to rearrange these because the manufacturer sometimes put those in there and wear it cause you have trouble to get them on so and then we're ready now we're going to open up our antiseptic because we're going to clean the surgical side before and then after we remove the staple before we place the steri-strip so we're gonna open up our alcohol prep sticks and you have three of those in there so we're going to take one of those and clean that area so we're going to take the swab and we're going to clean along the incision line on both sides this is going to help decrease infection now we're going to discard the swab and we're going to let this dry once the area is dry we're ready to remove our staples now our order said to remove all the staples and to do that what we're going to do is we're first going to remove every other staples starting with the second staple and while we're doing this is because we don't want this incision to open up prematurely so that helps hopefully decrease that chance of happening and if it does start to happen again what you want to do is you want to cover it up with a sterile covering and notify the physician so what we're gonna do is we're going to get our gauze and we're going to take it just lay it right there so we can easily drop each staple onto the gauze as we're removing it why are we doing this well these staples are small they can fall into a patient's bed on the floor and they're sharp and they can poke someone so keeping them on the goal this helps them stay collected and after you move each staple you want to count how many staples you've removed total and document that so take the staple remover and we're going to start at the second staple and what you're going to do if you can notice here the mouth of the staple remover we're gonna take this little part that's curved and placed that underneath the staple just like this and then we're going to depress the handle we're not going to pull up on the staple remover to help assist the staple to come out because this device will do it for you so we're just going to gently depress and see comes right out then we're gonna take our staple and drop it into our gauze now we're gonna repeat again go under the other staple depress pulls it out for you put in the gauze and then do the same thing with this we're gonna get a new swab and we're going to just clean those areas where we remove the staple and again just decreasing infection and we're gonna let this dry and we're going to apply steri-strips so we're gonna discard this swab so the area's dried and we have our steri-strips and we're going to apply a series strip to each area where we remove the staple now we have cutter steri-strips so that each side of the strip will extend about 3/4 of an inch from the incision they always follow your hospital protocols or whatever the surgeon instructs you to do with the length of the steri-strip now the thing you want to remember about series strips is whenever you're applying them you're gently just going to lay them on the incision you're not going to create tension or pull then or anything like that because these serious strips are strong and they could actually share the skin or tear the skin so as I do that you will see I'm not pulling or tugging on the strip to create tension and addition we want to keep about a one-eighth inch space in between our strips so with each package of steri-strips there tore bite there which helps you be able to access the strip easier so we're gonna take off our first strip and we're going to make sure we're lining it up where it has about 3/4 of an inch on each side and we're going to first lay it down on one side and gently just smooth it down and then not putting tension or anything like that lay it on the other side and smooth it down then we're going to get another strip and just repeat the same way okay again just long enough where we make sure we have enough on each side then we're going to place our last one on this round and then we'll remove the other staples again just lining everything up now we're going to remove the other staples just like how we did the other ones before so again just take the mouth of the staple remover can underneath the staple depress it pull it out and then dispose on the gauls and repeat we're paying attention to our wound making sure it's not opening up stay nice and close and it is here's our last one and we moved a total of seven staples so we'll document that now we're gonna clean the area with a new swab on where we remove the staples and then we'll let it dry and apply new steri-strips now the area's dried so we're going to go ahead and apply our steri-strips just like how we did before over where each staple was that we removed just line it up then we're going to apply our mix one just like we did the other one and these serious trips sometimes get stuck on your gloves so just be aware of that and we're going to apply our next one again just lining it up then we're going to apply our last one here at the end so we have our series strips on now you'll want to assess and think about you know is this area at risk for being rubbed up against something like friction or going to experience a lot of sweating here on the leg no not really but I've had patients who have had incisions in their groin and this area rubs up against the jeans it can get sweaty so you can apply a dressing over that and teach the patient how to change it regularly and monitor the site until the steri-strips fall off now the series strips will fall off all by themselves within about 10 day so tell the patient not to remove the strip's now after you've done that you'll want to dispose of your staples they're a sharps so they will go in the sharps box or dispose however your Hospital requires you to dispose of these and dolf your gloves and perform hand hygiene so that wraps up this video on how to remove surgical staples thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	Ostomy_Bag_Pouch_Change_Ostomy_Care_Nursing_Colostomy_Ileostomy_Bag_Change.txt	hey everyone is cereth register nurse sorry and calm and in this video i'm going to demonstrate how to change an ostomy pouching system in the previous video i did an in clicks review over colostomies and it we asked me so be sure to check out that video so first let's talk about the supplies you will need you will need a pouching system here I have a one piece pouching system and it has a skin barrier also called a wafer for land and it's attached to the pouch so it comes together and some people have the two-piece systems and the skin barrier and the pouch are separate and they snap on to each other whenever you're placing the system I'm also what you may need that I don't have right here you may need a barrier ring and the barrier ring will go around the stoma for extra protection in case the patient's having a lot of skin breakdown from still getting on to the skin and the skin barrier just isn't really protecting it so the barrier ring would go around the stoma and you would set the skin barrier on top it out and then also with this particular system we have a clip so be sure you get the clip if your patient needs one and some bags will velcro and that's how they close but this particular one uses a clip another thing you will need is a measuring card and these come in the kits usually and you will just be using it to measure the stoma because you will be cutting around your barrier what size the sum is so you can apply it to the skin so you'll need a pin you will also need some ostomies scissors these scissors are special and that they're curved and they have a blunt tip so you can achieve a circular cut instead of a jagged cut onto the barrier and of course you will need gloves you will need some wash cloths to clean the stoma and to dry it and it's how just to prevent any leakage of stool on to the patient first perform hand hygiene and then you will dawn gloves then give us how and place that on your patient to protect their gown and their skin from any soul by your change in the system and some things you want to keep in mind while doing this you'll want to change the system about every three to five days and you will empty the pouch whenever it becomes one-third to half way full and here as you can see this definitely needs to be empty but we're changing the system so it's good and and while you're doing this you want to allow the patient to help you as much as possible because they need to become independent in doing this because I'll be doing it whenever they go home now if you are a nurse who has a very sensitive nose and you're bothered by odorous smells and you need to be prepared for this because whenever you're changing an awesome e bag there are some pungent odors and to help with this you can maybe wear a mask I've even known some colleagues to put vapor rub around their nose before they go in to change the pouch to prevent the smell and so you can have a menthol smell and you notice you're changing it so just be aware of that and prepare beforehand so what we're going to do is we are going to remove the system and whenever you do this you want to do this when the gut is the least active because you don't want to be changing it and all the stool is just coming out on you because you're going to be making a huge mess so usually in the morning before breakfast before the patient eats is the best time or ask the patient because they'll know when there's some of hits off police but if this is a brand new summer that may not know so what you're going to do is you're just gently going to remove the adhesive because this is the one piece system and it has it's sticking to the skin and if you have difficulty removing it you can use some adhesive remover okay so we have that off and what you're going to do is just discard this appropriately then you're going to take your washcloth with warm water try not to use any soap have lotion powders creams or any alcohol containing products around the summer on the stoma because it caused some problems and before you do it you want to make sure you're looking around the stoma and looking for any skin breakdown like extreme redness now as you can see in this example there is redness around the stoma this is not good and what is happening is that stool is leaking probably under that skin barrier and getting on to the skin so whenever you're replacing the bear at the pouching system you may want to use a barrier ring and make sure that you're cutting the skin barrier should fit the stoma appropriately so it's not leaking onto the skin and what you'll do is you'll start around the skin and just gently clean the skin making sure to get any residue off and then you will clean your stoma and the stoma is not painful to the patient remember it's just the inside of the intestines flipped inside out so it's not painful for you to clean the stoma and make sure you have it all clean then you will Pat the area dry you want this to be very dry because if it's not dry your skin barrier is not going to stick to the skin hand firm if you have a patient who has a lot of hair on the abdomen because you know hair grows back you may want to trim the hair because number one it will stick better and when you remove next pouch change it's not going to wax or hair off which can be very painful so make sure you do that then discard your gloves and perform hand hygiene and put on a new pair of gloves now we're going to measure our stoma using our measuring card and the reason we want to do this is because we're going to be cutting a hole on our skin barrier and this is going to be going on the skin and we don't want to cut it too big we're still the leak to the skin or cut it too small where it will constrict the stoma so we want to measure appropriately and you'll take your measuring card you'll put it flush up against the skin and you want about a 1/8 inch an area around the stoma so this right here way too big and that is too big as well let's try this one and that one looks like it's perfect you have around the stoma about a 1/8 inch and look and see what it reads and it says 45 mil millimeters so we're going to match that up on our bag and trace it now some skin barriers are already labeled and they have a nice little outline of where your measurement is like here it already says 45 millimeters so I don't have to trace it but sometimes skin barriers don't have that so what you would have to do is put your measuring card over it and trace where you had your measurement so you was on the 45 millimeter and you would just create a circle on the barrier and then cut it out now I'm going to take my awesome scissors and I'm going to cut around the 45 millimeter mark and just go back through and make sure everything's nice and round and there's no jagged edges because you don't want this to wear on the stoma or on the patient's skin and we had it in looks good now let's put the new passing system on what I like to do is I like to snap my bag and make sure it's closed before I put it on just in case something leaks out so this is the clip like I said the beginning some velcro and these clips can be a little bit cumbersome to you so get familiar with how to use them what you do is you take the end of it the [ __ ] and you just simply fold it over the clip like so and then you just press that down and it snaps in place and you have your little clip on and then you will take the backing off of your skin barrier and if you're going to place a barrier ring now would be the time to do that so you place the barrier ring on around the stoma and then you would place your skin barrier your wafer flange on over the stoma and you want to make sure that it fits really good and that you smooth it around the sign so it's sticking to the skin appropriately and feels very good now some things you want to keep in mind if this is a new and awesome e for the patient you want to explain to them that they need to be checking their bags every so often because it can inflate with gas and they will need to do what's called burping their bag and to burp this particular bag you would take off this cliff and just let the air out now some awesome e bags do have a filter that allows the gas to escape and prevents an odorous smell from coming out so you just have to look to see what you have and just warn the patient that when they do relieve the gas from there that they may want to make sure that no one's around if someone's visiting them or something like that because it can produce an odorous smell and as the nurse you want to be checking this as well because our patients tend to wear those big bulky gowns and they can't include you being able to see the bags in plates so you'll have to pull the gown back and actually look at the bag so that is how you change an ostomy pouching system thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	Do_Not_Crush_Medications_Mnemonic_Crushing_Medications_for_Tube_Feeding.txt	hey everyone is Fairs rich Turner sorry and calm and in this video I wanna give you a pneumonic on how to remember the five categories of drugs that you do not crush as a nurse okay so what are those five categories we have the same relief extended release control release enteric-coated and long-acting so how are you going to remember those five categories okay remember this mnemonic seniors erroneously crush enteric-coated laxatives this mamani helps me remember the five categories because a lot of times requesting medications for patients who are older because they have difficulty swallowing so seniors and seniors tend to take laxatives because as you get older and the GI system they slows down they tend to have constipation ones so they tend to have laxatives lot and you would never crush enteric-coated laxatives like a dulcolax ec so each part of the word helps me remember the category so seniors s are abbreviation for senior is fr for sustained release all they may see SI for sustained action erroneously er stands for extended release you may see it as XR as well crush the CR seems for controlled release can sometimes eat a cd4 controlled delivery enteric-coated of course is for enteric coated EC and in laxatives LA is for long-acting you may sometimes see an abbreviate abbreviation XL so if you're ever wondering if this medication is a sustained action or release or something like that just look at the name of the drug because typically you'll have like a drug name that will say like get trough la we know that that's d'etre long-acting so we can't cross it the huddle is this pneumonic helps you out don't forget to check out my other videos on some mnemonics like the in forum video where I give you a new mom somehow remember categories of insulin along with the P&A sometimes and then the pneumonic on how to remember contact isolation thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	How_to_Split_a_Pill_in_Half_Cut_a_Pill_in_Half_Nursing_Medication_Adminstration.txt	hey everyone is sayers registered nurse ari and.com and in this video I want to show you how to split a pill so why do we split pills as nurses well sometimes a physician may order two milligrams of a medication however you are only supplied with four milligrams so that requires that you split the medication in half so in this video I'm going to demonstrate for you how to do that okay first what you want to do is you want to perform hand hygiene then you want to put on gloves and after you put on gloves you want to confirm that you have the right medication for the right patient's right route all those little checklists that you need to do then you want to get a pill splitter and this will help you achieve an accurate test because you want it to be even and if you don't get it even it will be in an accurate dose and these pill splitters have really sharp blades so never put your finger there because you will get cut and they have an area where you put the pill and it holds it steady for you as you cut it and you just do that and it'll snap it in half so we get our medication and you never want to split a pill with your fingernails your hands or some other object always use this and a lot of these medications come and with a scored line which help you achieve and even cut and if you're never sure of a medication can be cut in half always call pharmacy and asks in so we're going to take our peel and we're going to put it in our peel cutter now you're just going to close the lid of the cutter and push down and past it okay so say that you did this and the peel just disintegrates into all these little pieces and you don't know which parts which and always check with your Hospital protocol for whenever that happens places I've worked and they require that we send it back to pharmacy so they can credit the patient and that you start all over and also if this was a narcotic you would want to waste the medication another nurse and don't just throw it away so always follow your Hospital protocols with that as well okay so that wraps up this video on how to split appeal thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	Best_Veins_for_IV_Insertion_Drawing_Blood_Venipuncture_Tips_in_Nursing_Phlebotomy.txt	hey everyone xerath register nurse Orion common in this video I want to go over the veins that I love to use whenever I'm drawing blood or starting IVs so in order to be successful whenever you're drawing blood or starting IVs you really have to know a couple things number one you need to know the name of the vein that you're going to use and its location along with what can that vein actually handle for instance some veins can only handle about a 20 or a 22 gauge IV cannula versus some of them can handle 18 gages 16 gages so you really want to know which veins can do that now what means do I love to use whenever I'm drawing blood or starting IVs okay number one which I think the consensus is probably the same for this is the median cubital vein this is here in a moment you're gonna see a large vein right in the bend of the arm and it's perfect for drawing blood I also like to use accessories cephalic vein along with the median vein of the forearm and of course those hand mains the dorsal venous network so we have our tourniquet on up here on the top of the arm and I want to show you these veins so first I'm going to show you the media cubital vein and it is located right here and as you can see it's nice and large perfect for those blood draws and it arises out of your cephalic vein which is right here now this is our patient's left arm so on the right arm it's going to be the opposite but we have the cephalic vein right here and then over through here you can't see it as well as the bacilli and where the cephalic in the basilic start to connect as here at this median cubital vein now this main like I said great for blood draws you can use it for IVs the problem with it though is that it's in the band of this arm so as the patient is eating moving around getting up it's going to cause them a lot of pain and it can infiltrate and mess up the vein so unless it's like last resort you can't find any other veins on this arm or the patient needs like a CT PE protocol with an 18 gauge in one of those large veins you can go there use it and then just take it out and go somewhere else but I like to keep that available from blood next is accessory cephalic vein hence its name it comes coming off of that cephalic vein up through here now this is really one of my favorite veins to start IVs in number one it doesn't really roll it's easy to stabilize you're not gonna stick the needle in there and it rolls to the side don't you just love and that happens no not really and it's a relatively large vein so we can easily hold an 18 gauge now the thing is whenever you do use this for IVs because it starts up in here don't go in this Bend of this arm go down because it comes down and you can just stick your IV right there and the patient can move their arm bend it they're not going to have problems so this is really a good vein another vein I like to use is located on the forearm so we don't have a bend of an arm causing the patient and issues and it is the median vein of the forum also called the anti brachial vein and it is this vein right here it comes out of the palm of the hand now as you can tell this vein is not as big as the accessory cephalic and that media cubital vein so whenever you go to stick this for like an IV you want to probably pick like a 20 or 22 gage instead of an 18 gauge I have seen a limited amount of 18 gages in this fame because you know everyone's Anatomy is different some people have bigger veins there so you want to think about that another thing is that this these veins can run deep on patients who have a lot of sub-q fat on their forearms so you might be able to see the vein but whenever you go to stick it you don't immediately get that blood roll because the veins a little bit deeper so keep that in mind and that also is true for these veins up here on the bend of the arm as well and lastly the veins of the hand the dorsal venous network and these veins are great for drawing blood starting IVs but some things you want to remember okay first of all these veins like how this patient has their hand open if you go to stick a needle in this vein with the hand in this position this vein is just gonna roll see how that rolls like that what you'd want to do is you want to stabilize it so have them make like this like this and look that keeps that vein nice and still why you go with the needle to draw blood or advance a cannula now I have a whole video on how to prevent rolling veins if you need some more tips on that another thing to remember is that these veins sometimes they're really superficial on people like here they're nice and engorged and healthy-looking so they'll be great for an IV or drawing blood however on some patients who have a lot of sub-q fat on the hands they're hard to get to they're super small or like very fragile and they blow easy so always look at that look at the health of the vein before you try to stick it with a needle now when choosing a vein when you're starting an IV there are some things you want to keep in mind number one you want to ask yourself okay what's this patient here for are they just here for like a short stay because they need a quick procedure or they're just here for observation or they're gonna be hospitalized because they need treatment maybe they need some blood products or they're gonna have to get some vest can't type drugs like potassium think oh my son just to name a few so you want to select your site and your sizing your needle appropriately because for instance vancomycin is really hard on the veins and you want to use a large vein rather than a small vein next ask the patient where they prefer that you draw blood or start an IV because patients tend to be experts on where they know that you're going to be successful at getting that IV or drawing blood in some patients love to have IVs or blood rolls in their hands while some prefer that media cubital vein that's the only place they want blood drawn and that's the only place they want an IV so always ask your patient and lastly sometimes you're just gonna have to draw blood or start an IV where you can because if you're working with patients who have severe cardiac disease or renal disease they really don't have much to stick they're very very limited and where you can even get IV access and I've seen IVs and creative places like in the feet and the little fingers on the arms up in the neck anywhere you can get them so always keep that in mind and if that does happen and your patient is thinking lots of fluids IV drugs things like that you want to get with the physician you want to ask hey is this patient a candidate for a central line because I may benefit the patient a little bit better okay so those are some tips on choosing veins when starting IVs or drawing blood thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	Auscultation_of_Heart_Sounds_Assessing_Heart_Sounds_Listening_to_the_Heart_with_a_Stethoscope.txt	hey everyone it's s with register nurse rn.com and in this video I'm going to be going over how to osculate heart sounds what I want to be doing is I'm going to show you how to listen to Heart sounds on a real person I'm going to show you the anatomical sites how to identify S1 S2 talk about those S3 S4 and heart murmur those extra sounds you may hear but first let's cover the basics okay why do we listen to Heart sounds what is the purpose well one thing we want to make sure the rhythm is regular we want to count the rate but we're also one of the big things is that we're assessing how those heart valves are closing because whenever you are hearing S1 S2 those are valve valves closing S1 are your tricuspid and mitro valves closing and S2 is the sound of your aortic and P pulmonic valves closing and while you're listening to heart sound you'll be trying to distinguish am I hearing S1 S2 and then you're going to be positioning the patient in a little bit different positions and you're going to be listening for those extra heart sounds like S3 S4 and heart murmurs so first let's go over the anatomical sides here in a second you're going to see what it actually looks like whenever um you're looking at the anatomical sites on the chest but let me cover them real fast if you want you can write this down so you can remember it the key to help you identify these anat comical sites is to find the clavicle on the patient and then go down and find the angle of Lewis it's a joint little area and the second rib comes out from there and right below that we're going to start on the right side is the intercostal space and right there left I mean right of that border is the aortic valve and the aortic valve represents when it closes the sound of S2 that and the pulmonic valve normally close together so when they close together they're semi lunar valves you will hear S2 then right over on the left side in the same space second intercostal space you will find the pulmonic valve then down in the third space you will find herbs point and this is just an area where um you're separating the base from the Apex it's just the Midway point between those two areas and then you have the fourth intercostal space which right next to the border of that is the tricuspid valve you'll find that on the left side as well and the tricuspid and mitro valves when they close together simultaneously you will hear S1 and they are found in the base then you go down a little bit in the fifth intercostal space but midclavicular which is the Midway point of your clavicle and you will find the M mitro valve also called by cuspid valve and these are your Avo ventricular valves your AV valves and this is also where you you will hear the point of maximal impulse and also it's important to know the bottom part of the heart down in this area is the Apex the top part of this area is called the base you want to remember that now let's look and osculate these areas and see what they look like on a real person okay first I wanted to start out just showing you on the chest what where you're going to actually place your chest piece whenever you're listening to the what I like to do whenever I'm first starting out is either have the patient set up or lie down and I like to start in the aortic and work my way down remember the pneumonic all patients take medicine and herbs point is in between the pulmonic and the tricuspid and whenever your semi lunar valves are your aortic and pulmonic and when they close you hear S2 and so you're going to hear S2 the most at the base of the heart and then whenever you're hearing the tricuspid and mital which are your o ventricular valves which are AV valves you're hearing S1 whenever they close so let's use the chest piece and osculate okay whenever I'm beginning oscilation of the heart what I like to do is remove the clothing and um I like to have the patient set up you can also have them lay down and I listen with the diaphragm of my stethoscope first and then I'll switch to the bell and redo all the anatomical sites but I like to listen to diaphragm because you can hear S2 and S1 the best with this along on with your aortic and P pulmonic regurgitation murmurs so um I start at the aortic remember the pneumonic all patients take medicine and what I'm going to do is I'm listening for S1 and S2 I'm distinguishing them and I'm also listening for S1 splits or or S2 splits and this is just where the valves are not closing at the same time so you may hear a little bit of an extra noise so we're going to start in the aortic over here and what I'm hearing is love dub love dub and dub is louder because dub represents S2 and in the base of the heart you're going to hear S2 louder than how you would hear it down there then I'm just going to inch over here to the pulmonic and I hear the same thing I don't know any splitting S1 and S2 are closing at the same time no extra heart sounds then I'm going to inch down to herbs point this is just the halfway point between the base and the apex of the heart now I'm going to inch down to the tricuspid and this time I'm hearing love dub and love is louder because this is signifying more where you're going to hear S1 and love is represented by S1 and I hear that louder in this area and then I'm going to go over to the mital area midclavicular and this hearing the same thing love dub nice good Rhythm and what I'm going to do is I'm just going to switch over to my bell and I'm just going to repeat and what I'm really paying attention to is I'm listening for any type of murmur or those low pitch sounds you really can't hear S3 and S4 that great in this position that's why here in a second we're going to get on our left side and you hear that in the Apex but what I'm listening for is maybe any murmurs blowing swishing noise and I'm not hearing anything now one thing you may find hard whenever you are osculating is distinguishing S1 from S2 and some tips to help you with that again S2 is going to be louder here at the base and S1 is going to be louder here at the Apex so that can help you with that or if you're still having trouble you don't can't really differentiate um you can fill on the cored artery and listen at the apex of the heart and whenever you feel a pulsation and you feel you hear that noise you've identified S1 because the cowed pulsation and the sound signify S1 or if you have a patient on a bedside monitor you can look at your QRS complex in the r way the big spike whenever you see that that Spike and you hear the noise that is S1 so those are just some little tips on how you can differentiate between S1 and S2 now we've assisted the patient onto their left side and the whole purpose of doing this is majority of your heart is on your left side so whenever you turn them have them go there it pushes the heart over a little bit more just so you can hear those anatomical sites a little bit better and what we are interested in is the apex of the heart and we're going to be listening with the bell of our stethoscope because we're listening for low pitch noises and if the patient was going to have an S3 S4 or a mro stenosis murmur this is where we most likely hear it so what we're going to do is just find the midclavicular the fifth intercostal space we're going to just listen over there and we're listening for S3 or S4 and mmers and S3 is heard after S2 so again that's why you have to distinguish between S1 and S2 and S3 is going to sound like a love dub t love du T because it's heard after S2 S4 is going to be heard before S1 and it's going to sound like this t t t and a murmur of course is just that blowing swishing noise okay last what I like to do is I like to have the patient set up and lean forward and then have them exhale and I'm going to listen for what I'm looking for is murmur aortic and pulmonic murmur and I'm going to be listening at the aortic and the pulmonic SES with the diaphragm because it's good at picking up those murmur and what's happening is that the chest the heart behind the STM is just moving a little bit forward so we can hear those anatomical positions a little bit better and I'm listening for like a blowing a swishing noise and if one's present you'll want to grade that and here on your screen you'll see what the grading scale is for that one a grade one is hard to hear and it goes all the way up to six and this is the loudest you could literally lift your chest piece off the patient's chest like this and you could hear just the blowing and swishing noise you could also feel on the chest a thrill which is like a vibration on the skin okay so that is how you osculate heart sounds now be sure to check out my other video where I go in depth about these heart sounds I talk in great detail about them a card should be popping up so you can access that video so you can familiarize yourself with these heart sounds thank you so much for watching and please consider subscribing to this YouTube channel
Nursing_Skills_Videos	Carotid_Artery_Assessment_Jugular_Venous_Distention_Neck_Assessment_Nursing.txt	hey everyone its ears register nurse are en comm and in this video I'm going to be going over how to assess the neck specifically the carotid artery and looking at those neck vessels now this is typically done during the cardiac part whenever you're assessing a patient either from head to toe or you're just wanting to do a cardiac focus assessment and this is really important on those middle-aged or older adults who have a history of cardiac disease because what we're going to be doing whenever we're doing this assessment we are going to auscultate the carotid for brewery's which I'll go over that here in a second and we're going to palpate and we're going to fill on the carotid artery we want to know how the contour feels or the amplitude and we're going to grade it and then next we're going to assess those neck vessels for any bulging looking for fluid overload things like that so let's get started so first what we're going to do is we're going to auscultate the carotid artery and what we're listening for are breweries and that's what I talked about beginning the video and this is where you have a turbulence of blood flow and it will sound like a blowing or swishing sound and this is indicated of atherosclerosis narrowing and what you want to do is you want to use the bell of your stethoscope and we're going to listen in three different places to get a really good assessment we're going to listen at the angle of the jaw so find the jaw we're going to go right below it so we're going to go in there and then we're going to go mid cervical area which your cervical areas your neck and we're just going to go in the middle of it so we're going to go about right there and then we're going to go at the base of the neck and the key whenever you're doing this is you want to press lightly with the Bell of your stethoscope and whenever you're doing that you're just pressing lightly because you don't want to create a false bruit or compromised circulation by narrowing the arteries and what you want to do whenever you do this you want to have the patient take a breath and have them exhale and hold it for a second while you're listening so you're not hearing breath sounds and it won't mess with the sounds that you're hearing so we're going to listen real fast okay so first let's go at the angle of the jaw okay take a breath for me and exhale I take a breath for me and exhale again okay again okay and that sounded great and then you're just going to compare sides and repeat on the other side now let's palpate the carotid artery now whenever you're doing this you want to again compare sides like how you are whenever you're listening but what you're going to do is you're going to palpate lightly and what you want to use is your index finger and your middle finger finger do not use your thumb and avoid putting too much pressure on the carotid sinus which is up in this area because this has sensitive baroreceptors and especially in the older patients you're elderly patients you cause an vagal stimulation which will slow the heart rate down so do not do that it's generally good to palpate in the lower half of the neck to avoid that carotid sinus area and what you want to do is you want to locate the trachea and the sternocleidomastoid muscle and in that middle of that groove is where you will find that carotid artery and what you're looking for what you're actually feeling for is the amplitude and the contour you want it to feel smooth and rapid have a rapid upstroke and a slower down stroke and you want to grade it and a normal grading is two plus if it's really hard and bounding it would be three or four and if it's you can barely feel it it's diminished okay now what we're going to do is we are going to assess the vessels in the neck and we're particularly paying attention to those jugular veins and we're looking for increased central venous pressure and to do that what we're going to do is have the patient lay back supine on their back at a 45 degree angle and sometimes it helps to get an extra little light so you can shine it on the neck so you can see the vessels very good and what we're going to do first we're going to look for the sterno mastoid muscle which runs right here and we're looking for the external jugular which overlays this muscle and normally what you will see is that it's slightly flat again you might need your light or you may not even see anything at all which is okay and if you do see that this is bulging or very very large this is usually present in patients with congestive heart failure who have increased central venous pressure but here it's not some flat and you want it to look like that okay so that is how you assess the neck for your head-to-toe assessment now be sure to check out my other nursing assessment videos a link should be popping up so you can access that and thank you so much for watching
Nursing_Skills_Videos	How_to_Open_an_Ampule_How_to_Break_a_Glass_Ampoule_Nursing_Skill.txt	hey everyone its era thread sterner sorry and calm and today I'm going to show you how to break open a glass ampule so you'll want to confirm that you have the right medication and it's the right dose you're giving it at the right time via the right route to the right patient then you'll want to perform hand hygiene and gather your supplies you want to get your ampule that contains the medication you want to get an alcohol prep pad and some gauze then you want to inspect the ampule make sure it's not cracked anywhere the fluids not discolored or it has particles floating in it then you want to get all the liquid medication down there in the body of the ampule because sometimes it likes to collect here in the head and the neck so to get it down into the body what you want to do is just gently tap it and it will all go down into the body of the ampule then we want to take our alcohol prep pad and we're going to clean the neck of the ampule just to decrease contamination when we actually go to break it and we're going to let it dry completely once the ampule is dry we are ready to break it so we're going to take our cause and we're going to hold our ampule on a flat sturdy surface and sort of steady it and we're going to wrap the gauze around the head in the neck of the ampule like this and then when we go to break it we're going to break it in this motion like a snapping motion so whatever actually breaks the broken parts are away from our body so we don't get cut so take that gauze hold it around the head the neck and just snap it off then you're going to take this part of the head of the ampule and you're going to dispose of that in the sharps so you don't cut yourself along with the gauze thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	Popliteal_Pulse_Point_Palpation_Location_and_Assessment_Nursing.txt	hey everyone it's Sarah with register nurse rn.com and today I'm going to demonstrate how to find the poil pulse point and whenever you're assessing the pulse you will be looking at a few things one thing will be the rate how fast is it along with the strength and you'll be grading it on a scale zero to three with zero being absent one plus it's weak two plus it's normal and three plus it's bounding and then you'll want to look at the rhythm is it regular or irre and as you fill on the pulse you will be filling bilaterally to see if they're equal and this comes off the femoral artery and it's located behind the knee in these areas right here to find this pulse Point you're going to flex the knee and you're going to take both your hands put them behind the knee and you will find it at about the middle area of the ptil Fosso which is a diamond shaped pitted area behind the knee and this artery is pretty deep so helping to bend the knee will help you find that artery okay so that wraps up this video on how to check the ptil pulse point and don't forget to check out our other videos on how to check the other pulse points on the body thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	Deep_Tendon_Reflex_Examination_for_Nursing_Head_to_Toe_Assessment_of_Neuro_System.txt	hey everyone its sales ready Turner saurian calm and in this video I'm going to go over how to assess the deep tendon reflexes what I'm going to do for you in this video is I'm going to go over a short overview on how to assess a deep tendon reflexes as a nursing student or nurse then I'm going to show you how to individually assess the following deep tendon reflexes the bicep the triceps the brachioradialis the patellar and the Achilles and in order to do this you will need a reflex hammer and this is what a common reflex hammer looks like so you may be asking yourself what is the purpose why do we even assess the deep tendon reflexes well the D tendon reflexes help us to evaluate the lower motor neuron / v or fibers at specific levels in the body so what that means is say your intake your reflex hammer and you're in assess the brachioradialis tendon this tells us how the spinal nerve root C 5 - c 6 is working is it working appropriately or the triceps that you're going to hit on the tricep tendon it tells us how C 7 - c 8 is working and I'll go over those individually whenever I show you how to do them now as nurses whenever we are completing our assessments or head to toe assessments this is generally conducted during the neuro assessment part and depending on what specialty you work in as a nurse you will do this a lot more routinely than compared to other specialties now in nursing school you're going to have to learn how to do this probably get checked off by your professor in eliciting responses and those five deep tendon but as a nurse you're not going to do as commonly as you are like listening to the lungs every nurse is going to listen to all their patients lungs but not necessarily check their deep tendon reflexes branson's neuro settings and labor and delivery settings are probably going to use this skill a lot more than compared to a will say cardiac nurse because just to give you an idea in labor and delivery you have pregnant women and pregnant women are at risk for preeclampsia so one of the signs and symptoms so preeclampsia is hyperactive deep tendon reflexes so you're enough as a nurse you're going to be checking those deep tendon reflexes if that's happening and also on these women sometimes get magnesium sulfate because of these conditions and you want to be also checking those deep tendon reflexes now we talked about in the electrolyte videos you know with some of the electrolyte imbalances you can have diminished e tendon reflexes so say you are working on a med-surg floor and your patient has really low electrolyte levels you or high electrolyte levels you'd want to check those deep tendons to make sure you have a baseline compared to before so all of a sudden are diminished how were they whenever you assess them earlier so that is why we want to do this as nurses now and for documentation purposes whenever you do assess your deep tendon reflexes you use a grading scale now whenever you're checking deep tendon reflexes you need to practice this skill over and over because you have to understand what hyper reflexive versus hyporeflexia business because this skill is really subjective and it takes practice to understand it so let me go over how you would grade okay a four plus this is what is super really active this is the highest score you can get is considered hyperactive or clonus next is a three plus this is where it's brisker than normal it's considered hyperreflexia reflexive and two this is where you want the patient this is normal so hopefully your patient is scoring a two plus and one plus is diminished they're not very reflexive and you would consider this hypo reflexive and zero a score of zero means they're completely absent you get no response at all so now let's go over these five deep tendon reflexes and show you how to elicit response in one now it's elicit the responses of the deep tendon reflexes okay a lot of people whenever you're first starting out they have trouble eliciting a response so some tips whenever you are going to get that tendon to respond you want to make sure you swing your hammer nice and brisk not too slowly not too fast Lee make sure you move it back because you don't want the in order to rest on the tendon after you hit because you're not going to get the response you want also you want to make sure that you're hitting a tendon a lot of people whenever they're beginning they really don't know what they're hitting and what you're hitting is the tendon of that particular muscle so always make sure you're finding that and a good way to get to find that tendon is have to have the patient flex the muscle and you can feel a cord like area which is the tendon of where you're supposed to hit so the first one we're going to hit is the bicep tendon okay the first tendon reflex we're going to check is the bicep tendon and the bicep tendon is located below the bicep muscle in the antecubital fossa area if you're beginning out what you want to do is you want to find that tendon so to find the tendon we are going to have the patient flex the arm and flex the bicep and you will find it right in there and it feels like a cord like area and whenever you find it just put your thumb over it and this is going to elicit the response of c5 to c6 so I have my thumb over the tendon area and I'm going to have him relax the muscle and just drape his forearm over his lap and then I'm just going to hit with my hammer briskly and what you're looking for is contraction of the bicep and flexion of the forearm and we had a little bit of a response and there we go okay now let's do the tricep okay now we're going to check the tricep deep tendon and it is located on the back of the arm right above the elbow and if you're just new you're trying to find it what you can do is just extend the arm out and you can feel just right above the elbow that deep tendon so what you're going to do is you want this whole area to be relaxed so we're going to help the patient dangle their arm tell them to let it go limp and this is going to check c7 to c8 what we're going to do is we're going to take the hammer and just briskly hit on that deep tendon and what you'll see is extension of the forearm and a slight contraction of the tricep so here we go and seeing that seen that flex and that contract it one more time there we go now let's check our brachial brachioradialis okay we're going to check the brachioradialis deep tendon and this is checking c5 to c6 this can be one of the most difficult tendons applying because it's not as pronounced as the other tendon so what you want to do is you want to find the radial styloid process which is the nodule on the wrist and go about two to three centimeters above that how the patient turn where their thumbs up word relaxing the floor and resting it on their leg we're going to right above that and what we're looking for is for the hand to sup innate which the palm will turn out upward and you will see a little bit of flexion in the forearm okay you notice that right there okay now let's check our patellar and our Achilles okay to check the patellar reflex this is going to be looking at l2 to l4 and to find the tendon what you want to do is locate the kneecap and then go just a little bit right below it and the best way is to have the patient extend the leg out and then that tendon will pop out and that is where you're going to go okay after locating the tendon put your hand behind the knee just to support it have them relax and the tendon is right here and what you're going to do is you're just going to tap briskly and what we're looking for is this lower leg to extend outward so here we go and we've got a response there and one more time and got a really good response there okay now let's check the Achilles Achilles is going to look for l5 to s2 and this is best done having the patient dangle their feet just like whenever you're checking the patellar you usually want to check those two together and what you're going to do is you're going to Dorse dorsiflex the foot up like that and you will see the tendon which is located right above the heel and it's right there and what we're going to do is we're looking for the fit to plantar flex so it's going to go down like that too response but in order to do it you're going to dorsiflex the foot by supporting your hand underneath it and just tapping the bottom of it and flex melodeon okay so that is how you check the deep tendon reflexes please be sure to check out my other nursing skill videos and consider subscribing to this YouTube channel thank you so much for watching
Nursing_Skills_Videos	Upper_Lower_Extremities_Assessment_Nursing_Upper_Lower_Extremity_Examination.txt	this is cereth registered nurse Orion comm and in this video I'm going to demonstrate how to assess the upper and lower extremities and if you would like to watch a complete head-to-toe nursing assessment you can access this card up here in the corner or in the youtube description below now before performing this skill you'll want to perform hand hygiene provide privacy to the patient and tell them what you will be doing so let's get started so what we're gonna do is we're gonna inspect the extremities and we're looking for any lesions any redness swelling and this is a good time if they have a central line an IV that you look at that make sure it's not red does IV need to be changed that's that PICC line or central line need a dressing change assess that then you can palpate and what we're gonna do is we're going to palpate our pulse our radial artery so fill those bilaterally and they are two plus and they're equal then we're gonna check capillary refill and to do that we're just gonna press down on that nail bed and see how fast it comes back in it's less than two seconds then we're gonna check skin turgor but just pinch in the skin and see how fast it goes back and that was good then we're just gonna look at the range of the motion of the fingers and the hands look at these joints in the hands you see anything I'm normal like for instance like her badeen or both shards nodes which are found in osteo arthritis and ask the patient are you having any pain in your hands or anything like that no then you can palpate the brachial artery which is found in the bends of the arm and just fill those because that's another pull site and those are two plus and just as a side note if this was a patient that was getting dialysis and they had an AV fistula you would want to palpate that and feel for the thrill make sure that that is present up in that arm wherever their fistula is at then you want to test the muscle strength so what we're gonna do is I'm gonna have you squeeze my fingers as hard as you can okay okay that's really good then I'm gonna have you push up against my hands and I'm push up against your arms okay push okay very good okay and five plus normal strength then we're just going to test his put your hand underneath the elbow and just fill as you move the arm do you feel any grading crepitus of those joints one times an arthritis you can fill that and move that bilaterally another thing you want to do with the upper extremities is to check for a drift and what you will do is you'll have the patient hold out their arms and close their eyes hold it up for about ten seconds and you're looking for a drift like this so go ahead do that and close your eyes okay and we're assessing to see if this hand will drift upward and a lot of times if a patient has had a stroke okay you can put them in has had a stroke or something like that you will see a drift now we're going to assess the lower extremity so first of all we're gonna do is we're going to inspect we're gonna look at the color from the legs to the toes making sure it's nice and pink and here we see that being has a little bit of a tan line and we're looking at the hair growth as well you want to make sure there's normal hair hair growth because of PVD you will see hairless shiny thin legs in here we have excellent hair growth and also do you see any abnormal swelling just right off the bat before you've even touched the patient and look at the legs and the feet for any swelling redness swelling do you have any pain or anything in your legs anything like that and look in the at the joints make sure there's no redness on the joints cuz a lot of times with gout it likes to start out in the big toe so make sure that everything looks good and then on your diabetic patients make sure you look at the bottoms of their feet because these patients don't have the best feeling in their feet so their shoes could be wearing on them or they could have stepped on something and not even know it so inspect those feet make sure there's no ulcers or anything that like that that needs to be addressed also look at the toenails so the toenails look healthy or is there fungus are they missing toenails they have a really bad ingrown toenail so assess for that next you want a pouch your pulses will palpate the popliteal pulses which are behind the knee and those are about two plus they're equal bilaterally I'm just filling his legs they're nice and warm and I'm going to push over his tibia firmly and I'm seeing if there's any edema so push there and if there is a d-mail a lot of times whenever you push down it's like this hard light type gel it'll just separate and your finger will leave this indention in here we don't have any now we're going to palpate on the feet and we're gonna feel on the pulses and I'm gonna Don gloves perform hand hygiene and Don gloves and we're gonna feel on the pulses and this feet we're gonna feel on the posterior tibial tu+ really good and then we're gonna feel the dorsalis pedis which is on top of the foot 2 plus with that and if you can't ever find these because sometimes these are hard to find in patients you can get a Doppler if you have one on your floor next we want to check the capillary refill on his toes just like how we did with the fingers by pushing down in less than two seconds check the other one okay now I'm going to have him push against my hands push against my hands man okay good job now I'm gonna have you raise your legs against resistance good job now we're going to check the Babinski reflex and you can use your reflex hammer for this and use the end of it or you can use your finger if you don't have that and what we're gonna do is we're going to take this up through the ball of the foot and curve it and we're looking for the toes to curl in which would be a negative normal response so let's check that okay okay and that was normal then we're gonna Dolf her gloves and perform hand hygiene okay so that wraps up how to assess the upper and lower extremities and don't forget to check out the complete nurse head-to-toe assessment video thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	How_to_Apply_Remove_Transdermal_Patch_Fentanyl_Medication_Administration_for_Nursing_Students.txt	hey everyone and Sara Thresh Turner sorry and calm and in this video I'm going to demonstrate for you how to administer a transdermal patch specifically fentanyl so what is fentanyl finn'll is an analgesic it's used to treat pain and in this case patients who are using transdermal patches need them because they have chronic severe pain that isn't being managed with Pio medications so the transdermal patch will deliver this continuous amount of medication over a period of time so whenever you are going to be administering fentanyl you want to watch out for some things you want to check the patient's respiratory status because Pinto can cause respiratory depression so the rest or esata snore m'l also check their blood pressure because it can cause hypotension and assess the patient's pain rating and its location and see how well the patch is being effective and of course before you administer the fentanyl patch you want to do the patients v ride you need to ask yourself do I have the right patient do I have the right drug so you're going to look at the physicians order you're going to look at your past that you have in hand make sure says spindle you have the right dose and again comparing the doctors order with what you have in hand do you have are you administering it at the right time and is this the right route then you want to take take it a step further and look at your package make sure it hasn't been opened it's not damaged or been tampered with if it has you'll want to get a new one and you'll want to look at the expiration date each patch has an expiration data when it expires to make sure it's within date and since this is a teaching video we are using a demo dose so it's for stimulation only this patch does not contain fentanyl it's just to show you how to administer that and of course I always follow proper guidelines for administering medications and with the proper credentials so first what we're going to do is we're going to remove the old patch and you want to wash your hands and you want to dawn gloves and it's very important you use gloves for this because although this patch needs to be removed because it's expired time for a new dose it still contains medication you can get fentanyl on yourself so you want to protect yourself from getting residue from this patch and if you don't know where the patch is you can look in the Mahr to see where the last nurse who administered charted it should say where they put it typically they can be found on the upper arms the upper chest or the back sometimes on the flank so here on this patient it is on the upper arm so wash your hands run is on glove and we're going to remove it and anytime you're removing a fentanyl patch you need to have another nurse witness this with you especially the disposal part so what you're going to do is you're just going to gently remove the patch and then you're going to fold it together sticky side to sticky side like this and then dispose according to your hospitals protocol and because this is a narcotic again you're going to need another nurse to witness you disposing of this according to your institutions policy and then after you dispose it you can clean the side with some soap and water if it's left any residue because a lot of times from where it's sticky lint and old dead cells like to collect on the site so you can just take a washcloth and just with water and wipe the area all so after removing the old patch now it's time to sit on the new one and what you want to do is you want to open up the patch and you want to be careful not to use scissors if you can help it because you don't want to cut the patch that's inside of it so just tear the patch open like that and before you even start to take it off you need to date and time this and whenever you do this you want to make sure that you're doing it on the correct side because on this opposite side is the sticky side that you were removing the adhesive backing so it's the front to make sure you're on the right side so put the date sometimes your pins don't like to ride on these very well and put the time into your initials after day we night are going to dawn some more gloves because remember we want to protect ourselves from that fentanyl patch so we're putting on and whenever you're placing the pads you're not going to place it back on that same site you remove that you're going to rotate sites so again you can use the upper chest the other upper arm or the back and we're just going to use that other arm and you want to make sure that the skin is completely intact doesn't have a rash it's not broken and if it's really hairy in this area you'll want to clip the hairs don't shave but clip the hairs and put it there don't put it over any lotions or creams or anything like that and you may if the skin is dirty or oily you'll need to clean the skin with some water don't use soaps or anything with any type of lotions in it because it can affect how well this patch stick also another tip is if your patient is confused you don't want to put it somewhere where they can grab it off and it gets lost and maybe they have a child come to visit or someone they pick up that patch not knowing what it is I could get fentanyl on them so put it somewhere where they're not going to pick it off and try to pick an area that's not going to have a lot of friction on it or where it would easily come off as well so we're going to apply our patch over here and on most patches when you slightly bend them the backing will come off so gently take this part of the backing and just hold it up avoiding touching the back of the patch place that on the skin and the other piece will come off as well and here's to the skin and make sure that is on the skin there's no bubbles all the edges are down it's good to just hold it firmly down for about 20 to 30 seconds to make sure coming into contact with the skin correctly and if the patient gets sweaty or over time to some these are warm for several days you may have to place some tape over it or a Tegaderm works really well just to keep it in place then after you place your patch you'll doff your gloves and washer and you will chart it and it's very important you chart where you place the new patch so the nurse who's coming in after you and a couple days to change the patch will know where to find it easily so that is how you apply a transdermal fentanyl patch thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	Removing_Gloves_Properly_and_Safely_Technique_How_to_Remove_Gloves.txt	hey everyone stares register nurse re and calm and in this video I want to show you how to properly remove soiled gloves so our goal is to remove our gloves without getting the stuff that's on our glove on our skin or on our scrub so what we're going to do is we're going to take our thumb and our index finger and we're going to pinch the cup of one of the gloves then we're going to take the glove and kill it all and we're going to grasp the glove completely in our hand and close our hand tight prevent it from hanging down because we don't want to touch it then we're going to take our index finger of our unclogged hand and we're going to slide it underneath the cup of the glove and we're just going to peel it off and it fits it in a nice little Bunch and then we're going to throw it away and wash our hands thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	Cane_Nursing_Assistive_Devices_Mobility_NCLEX_Gait_Stairs.txt	hey everyone it's Sarah threads sterner sorry n comm and today we're going to talk about canes and as always after you get done watching this YouTube video you can access the free quiz that will test you on this assistive device so let's get started when you're studying these assistive devices for exams you want to make sure that you know those key concepts about those devices such as canes crutches and walkers and in our previous video we talked about crutches and in our next video we'll talk about walkers so for canes you want to make sure that you're familiar with how a cane should properly fit a patient how to actually walk with a cane how a patient should go up and down stairs with a cane and how they should get up from a chair or sit down in a chair with the cane therefore let's start with the proper fit how do you know as the nurse of this cane actually fits your patient well before a patient uses a cane for the first time it has to be adjusted most canes can be adjusted at the bottom by sliding the cane into location of where it should go to fit the patient but once the patient is holding the cane or once they have the cane standing beside their body how do you know it actually fits the patient well there's two ways you can tell the first way is that the top of the cane which is about this area here should be even with the great trochanter the great trochanter is a prominence of the femur so that's all the cane should rest about right there whenever the patient's standing up and has the cane beside of them or the top of the cane should be even with the wrist crease that is closest to the hand so those are two ways you can tell also whenever the patient is holding the cane the elbows should be flexed at a fifteen to thirty degree angle so those are some things that you can look for as the nurses tell you that this cane properly fits your patient now let's talk about how a patient should walk with the cane before a patient starts using their cane and practicing with it you want to make sure that they are wearing a gait belt for safety in addition you're going to stand on the patient's weak side in case they lose their balance so whenever they start using the cane they want to make sure that they're in the proper position the position that they want to be in they want to make sure that they are positioning the tip of the cane at least four inches from the side of the foot and they want to hold the cane on the strong side of the body so remember that the cane needs to be on the strong side very important concept to remember so how are they going to actually ambulate with this cane well to emulate with the cane what they're going to do is they're going to move the cane with the weak side together forward so they move the cane along with their weak side together and then they will move the strong side forward so how does a patient go up and down the stairs with a cane well the concept is the same like how we learned with crutches remember I told you to remember up equals good down equals bad and what we're referring to is the good leg going up first which would be the strong leg versus whenever we're going down this stairs it would be the bad leg that's going to go first so the weak side therefore how do we go up the stairs using that concept well what the patient wants to do is they want to hold the cane on that good slash strong side then they're going to move the good leg up onto the step and they're gonna put weight onto the cane and then move the cane and the bad leg up onto the step now to go down the stairs the patient again is going to hold the cane on the good side that strong side they're gonna move the cane down onto the step with the bad leg so the bad leg is going down then they're going to move the good leg down onto the step and lastly let's talk about how a patient will sit down and get up using a cane to sit down with a cane the patient is going to back up to the chair until he fills the chair with the back of the legs then the patient will allow the cane to rest on the side of the chair and place both hands on the chairs armrest and place weight on the hands while keeping the weak leg extended out and been the strong leg to sit down to sit up with the cane a patient's going to place the cane on the strong side and lean forward in the chair while keeping the weak leg slightly extended forward then the patient's going to push down on the canes hand grip and the chair armrest and then put weight on the strong leg and stand in position with the cane thank you so much for watching don't forget to take the free quiz and to subscribe to our channel for more videos
Nursing_Skills_Videos	How_to_Check_Vital_Signs_Checking_Vitals_Nursing_Assessment.txt	hey everyone it's Sarah with register nurse rn.com and in this video I'm going to be going over how to check Vital Signs what we're going to be doing is checking the six Vital Signs which are pain oxygen saturation temperature heart rate respirations and blood pressure and before you start what you want to do is you want to perform hand hygiene and you want to provide privacy to the patient and tell them what you're going to be doing because you're going to have to touch them in order to do this so let's get started the very first thing we want to do is we want to ask them if they are in pain so um whenever you do that you're going to have them rated on a scale of zero to 10 with zero being no pain at all and 10 being the worst pain they've ever had and if they do have pain ask them the quality what does it feel like and where it is at so hi Ben my name is Sarah and I am your nurse and I'm going to be getting your Vital Signs and I performed hand hygiene and very first thing I want to do is I want to ask you what your pain rating is are you having any pain rate on a scale of0 to 10 yes pain in my shoulder and it's a three okay and what is it feel like it's just a sharp pain when I raise my arm okay so you're having a pain of three and it's in your left arm and it's sharp yes okay now I'm going to get your temperature there's several ways you can take a temperature every facility has a different system set up so use what they have but you can take it orally you can take it axillary you can take it tanic in the ear or you can take it temporally or rect um rect is the preferred route usually on your pediatric patients but in adult patients normally we do it orally some things to keep in mind though axillary and temporally the readings are going to run about one degree lower than oral and for tanic and rectal temperatures it's going to usually run about one degree higher than your oral reading so we are going to check this orally and what we're going to do turn your thumb Omer on make sure you're using the proper um sleeves if you have any sleeves for it clean it everything like that follow your hospital protocols and have the patient lift up their tongue and put the probe underneath the tongue and have them close the mouth with the tongue over the probe and hold it there until it beeps a normal temperature is about 97° fit to 99° F okay and take the thermometer out and read it and his temperature is 98.2 and then clean it properly per Hospital protocol now I'm going to take his oxygen saturation every system has different ways of how they measure it um different devices this is a little portable device and what you do is you put this on the nail bed of the finger it has some red lights in there and those red lights read through the nail bed the oxygen saturation a normal oxygen saturation O2 sat as you may hear and the hospital setting is 95% to 100% so let's see what his is um put this on the index finger of the nail bed and then just look for the reading okay here you can see that his oxygen saturation is 98% that is the reading on the top it's read as spo2 and then on the bottom you will see his heart rate which is 64 but here in a second I'm going to show you how to actually count the heart rate using the radial artery okay now we are going to count the heart rate and respirations generally I like to do this together um while I'm counting the heart rate I count that for 30 seconds if it's regular and then the next 30 seconds I count the respirations which I look at the rise and the fall of the chest and that equals one breath um generally if you tell a patient you're going to count their respirations they change the rate of breathing so it's good to conglomerate those two together so you can get a more accurate reading so what we're going to do is we're going to first count the heart rate and to do that you can use several different sites typically people use the radial which is right here right below the where the radius bone is and the groove right there or you can use the brachial artery which is in the bend of the arm where the anticubital fosset area is or you can use the cored but here we're going to use the radial so what I'm going to do is I'm going to use my two I'm going to use my index finger and my middle finger don't use your thumb because you can feel a pulse in your thumb so use those two fingers and just put it over in the groove of where the styloid processes and the radial artery and feel that and count for 30 seconds if it's regular if it's irregular count for 1 minute and a normal pulse rate in an adult is 60 to 100 beats per minute okay the heart rate I got 60 and his respiration were 16 now we are going to get his blood pressure now whenever you're getting blood pressure you want to make sure that you get the right size cuff in most settings they have the automatic blood pressure cuffs where you don't have to blow it up yourself so you're really blessed with that but a lot of times you may have to learn how to do a manual one now my previous video and a card should be popping up I go over the two-step method if that's how you're being instructed but in this video we're going to go over the one step blood pressure of how to obtain it manually so what we're going to do we are going to palpate the brachial artery this is in the bend of the arm and make sure you ask the patient which arm you can take their blood pressure in because you don't want to take it in uh arms with if they've had blood clots or they have a shunts things like that so you want to make sure you have the right arm and what you're going to do is you're just going to have them extend the arm out and you're going to palpate the brachial artery this is found in the anticubital faucet area and the bend of the arm towards this area and um extending the arm out helps that pulse really pop out at you and just f that and we feel about right here because what you're going to do on your blood pressure cuff you have these little arrows and it says left arm right arm and this is his left arm so we're going to make sure that we put this Arrow about 1 to two inches above that artery so let's slide it up and then make sure our cff it properly so we're putting that Arrow about one to two inches above where I felt the brachial artery and going to just put this on here snugly and to make sure you have the right blood pressure cuff take about two fingers and slide it underneath the cuff and make sure it fits snugly not too tight not too loose because if you don't fit it correctly you could get in inaccurate blood pressures okay so we have that there and put your little spigon monometer somewhere where you can see it because that is where you're going to be finding your blood pressure so put our stethoscope in her ears and you're going to use the diaphragm of your stethoscope and you're just going to place it over where you have heard that brachial artery and then you're going to blow the cuff up to about 180 to 200 mm of mercury or until you don't hear that braak your artery anymore okay we're blowing it up to about 200 mm of mercury okay we're there and now we're going to let the needle drop about 2 to 3 mm of mercury slowly not too fast not too slow and we're listening for that first sound and that first s sound will be our top number of our blood pressure which is our systolic so I haven't heard it yet and I'll let you know whenever I hear it okay I heard it at about the 114 Mark now we're listening for whenever it stops and whenever it stops that's our diastolic okay it stopped right at 65 so his blood pressure is 114 over 65 so that is how you check bottle signs now whenever you're done remember to let the patient know what their bottle signs were and um do hand hygiene and clean your equipment before you go to the next patient so be sure to check out all my other videos on nursing skills and thank you so much for watching
Nursing_Skills_Videos	Lung_Sounds_Abnormal_Crackles_Rales_Wheezes_Rhonchi_Stridor_Pleural_Friction_Rub_Breath_Sounds.txt	hey everyone this is Sarah with register nurse rn.com in this video I'm going to let you listen to six abnormal lung sounds which will include wheezes Strider crackles and plor friction rub this will be part of a review series of the lungs also be sure to check out the video on normal lung sounds as well as a lecture and skills demonstration which is part of this series so let's get started [Music] [Music] m for okay thank you so much for watching remember you can go to our channel to find more videos on nursing skills inlex prep and more so please subscribe and share this video with others
Nursing_Skills_Videos	Eye_Drop_Administration_Nursing_Instill_Eye_Drops_Punctal_Occlusion_for_Glaucoma.txt	hey everyone it's Sarah register nurse rn.com and in this video I want to show you how to administer eye drops so the first thing what you want to do is you want to perform hand hygiene and you want to wash your hands so we're going to do that real fast and use some hand sanitizer and we're going to Dawn gloves and after we do that we are going to first assess the eye and what we're looking for we're looking for any excessive redness or drainage and a lot of times patients um they may be getting eye drops because they have an eye infection and if you're giving it sometimes usually in the morning that eye may be matted up and closed or have exu day around it so what you want to do is you want to clean it so to remove any excessive drainage or Crush you're going to take a washcloth with warm water and you're gently going to start at the inner part of the eye and you're going to work your way outward just gently removing it and then you're going to change your gloves and wash your hands and put a new pair of gloves on now after you've wiped the eye what you want to do is you want to do the patients right you want to make sure you have the right patient the right medication the right dose the right route and the right time and when you're ad ministering eye drops you want to make sure that you really pay attention to which eye drops go in which eye because sometimes patients have more than one type of eye drop they're using and how many drops goes in each eye so be aware of that then you need to position your patient so what you want to do is you want to have your patient tilt their head back and look at the ceiling now if your patient cannot do this they're in the bed um what you can do is you can lay the bed back a little bit and reposition their head with a pillow to assist them in being able to get in this position because we want to do this so whenever we instill the drops they don't easily leak out and they can go really easy into that conjunctival sack now you're going to take your non-dominant hand and you can use e either your thumb or your finger and get a piece of tissue paper to collect any excessive drainage and you're going to rest that on the cheekbone and what you're going to do is you're going to gently pull down on that lower lid to expose the conjunc sack because this is where we will instill the drops we're not going to instill the drops on the eyeball over the cornea because that is um a sensitive area but in this sack right here at the lower lid then you're going to take your dominant hand and just rest it a little bit above the eyebrow and you're going to um take the eye drops put them in between your thumb and your index finger and gently tilt them over and we're going to in still in the conjunc sac the prescribed amount of drops and you want to be careful not to get this tip in the eye and contaminate the tip of it so we're going to instill the two drops one two okay and then the patient's going to close their eyes and tell them to roll their eyeball around to coat the eye with the medication and dab the eye with the tissue paper to collect any excessive drainage now sometimes after you administer eye drops you may have to perform punctal occlusion and this is usually done whenever you're administering a medication for a patient who has glaucoma and it's used for medications like the beta blockers a popular one like Timol or any of your medications that end in o LOL or Alpha ad energic Agonist and to do this you will take the finger and you will put it to the bridge of the nose over the punct and hold it firmly but gently to block the solution from going down into the nasol laral duck because if that happens the medication we want to stay in the eye we don't want it to go down through the throat and enter into the systemic circulation because one um the patient isn't getting the full dose in the eye and second it can cause signs and symptoms that beta blockers can cause throughout the body and we don't want that to happen so let me show you how to do punctal occlusion so after uling the drops have the patient close the eye and then just take your finger and you're going to hold at the bridge of the nose where the punctum are and you're going to hold firmly but gently for two to three minutes and have the patient just move their eye around to coat their eye with the solution but just keep those Ducks uded so that is how you administer eye drops thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	Hard_to_Find_Veins_Blood_Draw_IVs_Venipuncture_Nursing_Phlebotomy.txt	this is here with register nurse re and calm and in this video I want to go over some tips on how to deal with those patients who have difficult veins as a nurse it is inevitable that on a daily basis you're going to encounter those patients that are difficult to draw blood from or start IVs and there's various reasons that this happened some reasons could be your patients just severely dehydrated so those veins are not popping up or they have severe cardiac or renal disease so what I want to do is I want to give you some tips on how to deal with those things the first thing what you want to keep in mind is that usually on most patients they're gonna have something for you to use for drawing blood or starting an IV even though it doesn't look like it initially or the patient tells you no I'm a really hard stick but usually most patients will have something that you can access so with these patients you're really going to be going by palpation of that vein rather than actually seeing it like you know on patients who are healthy they do a lot of cardiovascular exercise those veins are nice and superficial at the surface and once you apply that tourniquet they just pop up there but on patients who have bad veins that doesn't happen you'll put the tourniquet on there and those veins really just don't pop up as great as someone who has those nice and gorgeous aims so you have to know number one the location of your veins and in the adult they're usually the same across the board like the median cubital the accessor area cephalic you're just going to find those in the same location so be familiar where you can find these veins second what you have to do is you have to practice filling veins and veins feel very unique compared to other structures in the arm so what I like to do is I like to take my index finger and I like to start filling all throughout my arm the front the back everywhere I can especially on these patients whose veins just aren't popping up because veins have a very unique feel in that they're squishy but bouncy and once you get the feel of what they feel like you will never confuse it for other structures and arm but whenever you're first starting out it's very foreign because you know I remember starting out and filling on an arm I'm like what is that is that a vein and and my preceptor be like no that's like a tendon or a ligament I'll be like oh okay so what you can do what worked for me is that I would take a tourniquet apply it to my own arm and I would feel my veins or I would use my husband or a friend or a family member who ever was healthy who had great veins and I would just feel throughout the arm and get comfortable with how that vein would feel next you want to use two things you want to use a tourniquet and gravity both of these things are going to benefit you greatly when trying to start a V s or draw blood on these people who have difficult veins because it's like this same concept with the water hose once you kink that water hose and put a lot of pressure on it what is that water hose start to do it starts to swell that's what these veins are going to do because we need that once we go in there to palpate to fill that swollen squishy bouncy vein so we can stick it so what I like to do is I like to put the tourniquet on the upper arm not too tight but not too loose just alright have a patient drop the arm by the side and pump out this just open and close that hand over and over and this gives a time for that blood to pull in that lower part of the arm so you can increase your chances of getting a vein and the thing is it's really amazing how taking the time to just do those two simple things really does work now I worked in a stress lab for many years and right off the bat whenever the patient comes to you for the tests they have to have an IV and I would get a lot of patients who would tell me you know people hardly ever can stick me successfully it takes multiple tries I just don't have anything and you know looking at their arm initially they really didn't have anything but once I apply that tourniquet use gravity for my benefit it was just simply amazing what actually popped up and was available for me to use they would shock me and shock the patient and we usually get it first try and the patient be like wow that never happens so always try to use those two simple things to help you with drawing blood and getting IVs and lastly you know sometimes you can put the tourniquet on you can use gravity all day you can palpate all day stick the patient you're just not getting anything and this is where you need to use those outside resources there's vane lots available maybe your unit has one that can help illuminate the vein so you can easily find it or in other situations depending on what you have available at your facility you can have ultrasound ultrasound the arm find that vein and stick them for IVs and places I've worked we've had that and I'll share a story with you we had a patient coming in for a stress test and they were like you know I need a central line to have this test and you know going and getting a central line is expensive for the patient there's risk of infection and all these issues just the HAP like this for our tests really an IV is the best route so we called ultrasound they came up they ultrasound the arm and they found a vein it was deep down in the bend of the arm but there was a vein there starting IV got it person had their stress test and we saved them from having to get a central line and that was great so always look and see what your Hospital has available for those situations that do arise but if you do have a patient you know you're just not getting anything on them they're gonna be there because they're really sick they need a lot of treatment with IV fluids you probably want to get with the physician explain what's going on because I patient it may be a candidate for a central line and do that and said okay so that is what I'd like to do whenever I have those patients who have those difficult veins and I hope those tips help you out and be sure to check out my other videos on how to start IVs and tips on how to deal with those rolling veins and things like that thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	Blood_Transfusion_Procedure_Nursing_Reaction_Types_Complications_HemolyticFebrile_NCLEX.txt	this is cereth registered nurse re and calm and in this video i'm going to be going over blood transfusions which will include the nurses role and as always I've been to this YouTube video you can access the free quiz that will test you on this procedure so let's get started what is a blood transfusion it is where as a nurse we will transfuse a patient who is low on red blood cells with new red blood cells via a venous access of some type now this is most commonly done through donated red blood cells so a patient needs them some ones donated them to the blood bank and as a nurse we will hang a bag of red blood cells for the patient and transfuse them to replace those low red blood cells which answers the question why would a patient need a blood transfusion because they're low on red blood cells and what can cause a person to be low on red blood cells well number one blood loss and they've had some type of surgery or trauma they can lose a lot of blood so we have to replenish them with red blood cells or they're anemic they can have anemia so bad that they need blood transfusions at some point like because they don't produce enough red blood cells and this can happen in conditions like renal failure cancers just to name a few because the body is not producing enough substances to produce or maintain those red blood cells now what's the importance of red blood cells they are very vital for our survival and how our body works so in other words our body can't function very well without them so what a red blood cells do with the help of hemoglobin it carries oxygen it receives from the lungs throughout our body in addition it removes carbon dioxide co2 and it will take that take it to the lungs so the lungs can't exhale it therefore whenever your patient is low on red blood cells they're going to have some side symptoms that can present especially if they're really low they'll be very pail I've seen patients they literally look white as a sheet before the transfusion and then after the transfusion I've told all my patients man you look like you got a tan because they their skin color is back to where it should be so it's really interesting you transfuse blood make sure you look at that another thing is they can feel very fatigued they can be short of breath any activity they're just like really wore out and they can have an increased heart rate be tachycardic because that heart is trying to pump that blood because consents of the oxygen is low so it's like I got to get more blood everywhere else so it can overwork itself so when is a patient transfuse well this really depends depends on what's going on with the patient their vital signs how are they tolerating that low blood level and recent guidelines by the American Association of blood banks recommends transfusing blood when hemoglobin levels fall to 7 to 8 grams per deciliter so then what is a normal hemoglobin level well depends if you're male or female Mel's it's 14 to 18 grams per deciliter in females it's 12 to 16 grams per deciliter now let's look at the nurses roll with transfusing blood now transfusing blood is very common in the hospital setting and you not only need to know this for NCLEX your Nursing lecture exams but you need to know it for the job okay the first thing before a patient is even transfused is a lot of prep work that is super important and essential because our prep work helps prevent transfusion reactions so we want to make sure we follow exactly what we need to and hospitals have in place protocols that whenever you become a new grad you start working you want to read over there protocol and make sure that you follow it exactly now most hospitals require that you're a registered nurse in order to transfuse the blood so again follow your Hospital protocol with that so let's say you got an order for patient to be transfused with two units of packed red blood cells what's the very first thing that's going to be done the patient is going to be type and cross-match either you'll be drawing the blood or your phlebotomist will be drawing it and this is the part where you've got to pay special attention to everything from what you write down to how you identify the patient you place the blood band everything must be done perfectly because we don't want do too like some type of clerical error to cause a transfusion reaction because someone messed up which tends to be the most common reason for a transfusion reaction so always take care whenever this has to be done next send the blood labs going to type in blood bank will match the blood with a donor and we'll prepare the amount of blood that you're going to need now it's also important that as the nurse you know the patient's compatibility and their Rh factors what blood they can receive what blood they can't receive and I have a whole video where I went over that in depth with you and you can access that and take a quiz that can test you on that but just a quick review who is the universal donor who can donate to all types that's oh now who is the universal recipient they can take from everyone that's a B next you'll want to get informed consent tell the patient what they're going to be receiving assess their understanding of it also this is a good time to ask about their allergies and if they have received any blood transfusions in the past and if they have how many because if they have received a lot of blood transfusions in the past they're at risk for febrile nan hemolytic transfusion reaction where their body has just built up these antibodies from all those previous transfusions and they can start running a fever and things like that so a lot of times physicians like to pre medicate them and you'll want to let the physician know if they do have a history of that and sometimes they're pre-medicated with benadryl or Tylenol acetaminophen before hand orally or when about 30 minutes before you start the transfusion and that will help prevent that also look at the health status of your patient are you giving a patient who is in fluid overload or congestive heart failure has renal failure and but they really need blood you need to be looking at that because they may be at risk for circulatory overload where you can put them in fluid overload because you're putting all this blood inside of their body so a lot of times physicians may order lasix some type of loop diuretic before the transfusion or in between the units or after the transfusion so you want to be aware of that as well next you want to make sure your patient has IV access we have to get this blood in then and you typically want an 18-gauge or larger IV site some hospitals again it's varies on protocols they'll allow you to transfuse through a twenty engage and why is that well as those red blood cells are shooting through they're going into the system if that cannula is not large enough those red blood cells can break open they can lysis and you're just breaking them up and they're not really going into the patient's body so you want good IV access another thing you really want to consider is you know it takes anywhere between two to four hours for a unit of blood to transfuse well if you have to hang some antibiotics on this patient or they're gonna need some IV drugs you can't use that IV access that is being used for the blood transfusion so you need a second access so keep that in mind it's always good to just have a good two sites while a patient's needing blood next is supplies whenever you transfuse blood you use special tubing which is called Y tubing with an inline filter which helps filter some of those substances out of the blood before it actually goes to the patient and keep in mind again it depends on hospital protocol a lot of protocols say only one set of Y tubing per unit that you transfuse so you'll have to you're going to transfuse the patient with more than one unit multiple sets or some hospitals say it's only good for four hours so keep that in mind when you need to change your filter your tubing next you'll want to grab a bag of normal saline 0.9 percent of normal saline this is the only only solution you ever use whenever transfusing blood you never use any met other medications or any other fluids only saline remember that because say you gave it with a dextrose containing solution dextrose and red blood cells don't get along it can cause them to clump up together so only 0.9% normal saline and we will be using that saline to prime that why tubing whip and then to flush that tubing with afterwards once the blood is done transfusing and in addition just whenever everything's wrapped up you're ready to take that tubing down you'll need to get a red biohazard bag to dispose of it properly again fall your Hospital protocols for holiday dispose of their blood products but um you never ever put it in the regular trash now let's talk about transfusing okay done all your prep work the blood bank calls you and says hey your blood is ready let us know whenever you're ready for us to send it to you because they're keeping it refrigerated for you so some key things you want to remember you will be giving one unit at a time patient needs two units you're gonna give one unit now and then whenever that's done call blood bank and say send me the other unit and then they'll send it to you so one at a time and from the time that the person brings you your blood or you go and collect it you need to start transfusing no later than within twenty to thirty minutes and it needs to be done that unit needs to be done within two to four hours so why is this well from the time that that blood bag leaves the refrigerator it needs to be in the patient's body within no more than four hours because there's a risk of it developing bacteria and we can give the patient septicemia so you want to transfuse it in that time frame also another thing you just want to consider one throw this out there is notify the blood bank that you're ready for the blood when you're ready because you need to get this in and as a nurse you get your blood so you have mission discharge you have a patient that is just not doing good that 30 minutes flies fast and you have this blood and you haven't even given it and now it's like 35 minutes later well you can't give it you're gonna have to send it back to the blood bank and blood is expensive so notify the blood bank whenever you're ready to start that transfusion now blood wormers blood warmers can be used if the patient needs large amounts of blood quickly and they're at risk for experiencing hypothermic response so you're not gonna warm the blood up by using a microwave or anything like that you want to use a special device if need be next before you even start the transfusion you're gonna be doing this verification process so as the nurse you're gonna be getting another nurse it's usually to our ends cuz ourian's are usually the ones who can transfuse and then you're gonna do another verification process with another orient and you're going to be looking at the following things before the transfusion together you're going to be verifying that physician's order you're going to be looking at the patient's identification versus the blood banks information making sure everything matches up perfectly you're gonna look look at the patient's blood type versus the donors tie and the Rh factor you're gonna make sure that they're compatible next you're gonna look at the expiration day on the blood make sure it's not expired you're gonna look at the blood make sure it doesn't have any clots are abnormal substances in the blood or it's damaged in any way and everything must match perfectly and if there's a discrepancy you'll need to notify the blood bank immediately and just from personal experience this has happened with one of my patients I was doing the whole verification process with another nurse and we were looking at the blood bag looking at the patient's ID band looking up and there was one letter that they had did a clerical error on so I had to send the blood and we had to go through the whole process again so this does happen so always make sure you verify everything also before transfusing you're going to be getting baseline vital signs which is going to include the temperature the blood pressure respirations and heart rate you want to make sure those are within normal limits and especially that temperature if you have a temperature greater than a hundred degrees Fahrenheit you'll want to notify the physician and make sure they just want to still proceed with the blood transfusion then again before you actually transfuse you want to explain to your patient if they're alert and oriented they can talk to you what you're about to do and for them to notify and report to you if they feel any of the signs and symptoms I'm fixing to describe because it can be a transfusion reaction like sweating chills chest pain itching short of breath headache backache or nausea and vomiting and if this happens you'll immediately want to stop the transfusion okay now it's actually time to start the transfusion so you're gonna have your blood ready hung and it's going to be controlled by an infusion pump which will deliver it to the patient and you want to start the transfusion slowly about two milliliters per minute for those first 15 minutes in addition you want to stay with that patient at their bedside looking at them monitoring them for those first 15 minutes and why is that why are you doing that well you're writing up slowly because we want to minimize the amount of blood that the patient's going to receive in case they do have a possible transfusion reaction we can turn that blood off and we want to stay with them during the first 15 minutes because that's when most transfusion reactions occur so we're gonna be watching their bottle signs throughout so you're gonna stay with them after five minutes of starting the transfusion you're going to get bottle signs and again this is depending on your hospital protocol and then after that you'll get them at 15 minutes after the transfusion has started and if the patient's okay they're tolerating it well this is the time that you can increase the rate and remember you want to make sure that that unit that bag of blood goes in within four hours no more than four hours then you'll get it to 30 minutes again and then hourly until done and then one hour after the transfusion and throughout that blood transfusion you're going to be monitoring them for of course a transfusion reaction and transfusion reaction that word is like an umbrella term for a lot of different reactions the patient can have and here in a moment we're gonna go in-depth over those but it's where the recipient that patients immune system is interacting with the donors blood so you can have a hemolytic transfusion reaction where what's happening is that the patient's blood and the donors blood are not compatible and immune system is attacking that donors red blood cells that they're receiving and this is dangerous it can lead to death also allergic they can have a febrile which is non hemolytic or GVHD which is graft-versus-host disease and have an Astra cry this because this tends to happen days two weeks after a blood transfusion it's rare and it's deadly if it does happen and in addition you want to monitor your patient for what's called circulatory overload and which patients do you think would be at risk for this think of any patient that's at risk for whenever you put extra fluid volume in their blood they'll have trouble with it like patients who have heart problems like congestive heart failure their heart muscles weak and you just throw that extra fluid in it it can't do well with it so the fluid starts backing up into the lungs and into the tissues they have breathing trouble things like that also patients who have renal failure you know these patients need blood but their risk for being able to tolerate all that fluid going in there so you want to keep that in mind you do have to transfuse these patients who have that in addition we want to monitor them for septicemia now it's in a way on how we can remember all those big signs and symptoms that your patient may be having a transfusion reaction so to help us to remember those major signs and symptoms let's remember the word reaction okay r4 rash they may have a rash or fives efore elevated temperature and what you want to do is you want to look at that baseline temperature and you need to ask yourself how's it increase and if you're measuring in Fahrenheit has it increased 1.8 degrees or if you're recording in Celsius has it increase one degree from baseline if so think transfusion reaction a 418 is your patient saying I have a backache all of a sudden or I'm having chest pain or my head is hurting that's a red flag c-4 chills t4 tachycardia especially if it's really increased from baseline I for increased respirations same thing with that increase from baseline oh for all glory so you really want to be looking at your patients urinary output during this blood transfusion and after are they putting out low or are they just putting out no urine at all are they an Urich and then look at the color what does it look like are they experiencing a condition called hemoglobinuria where there's free hemoglobin in the urine it will have like this purplish color so watch the earring closely and then in for nausea GI issues like diarrhea then when the transfusions done your patients tolerated it well you'll want to flush that remaining blood out of that line with that saline that's hanging on that Y tubing then dispose of your tubing properly and then collect post vital signs one hour after that transfusion now let's take a closer look at those various transfusion reactions and then talk about what you're going to do as a nurse if your patient does have a transfusion reaction during a blood transfusion okay first hemolytic this is where the immune system is killing the donors red blood cells so what's happened antibodies in the recipients blood match that antigens on the donors blood cells so hence they've been missed tight and this can lead to di C and renal failure and even death and a lot of times what's going to happen is you're gonna see a fever chills anxiety back pain chest pain hemoglobinuria where you have that purplish look to the urine also they can be tachycardic and have a low blood pressure another type is called allergic and this is where the recipients immune system is reacting to the proteins found in the donors blood leading to like rashes hives and itching and it can actually progress to in a flexus and the patient with this can have hives rashes respiratory issues like wheezing oral swelling things that you expect whenever someone's having an anaphylactic reaction to something another type is febrile and this is non hemolytic so you don't have the breaking up of those red blood cells like you did in hemolytic but this is where the recipients white blood cells or reacting with the donors white blood cells and this causes the body to build antibodies so you can see that increased temperature like one degree Celsius or one point eight degree in Fahrenheit from the baseline and this is actually the most common transfusion reaction that you tend to see especially in patients who have received blood in the past because their body has created these antibodies so that's what you want to ask them have you received a lot of transfusions before and you can see chills headache increase heart rate and fever with that and another transfusion reaction you can have is the GVHD the graft-versus-host disease and again like I said this is rare but it's deadly and it tends to occur days to weeks after the transfusion so this is where the donors T lymphocytes cause an immune response in the recipient but actually in grasping in the marrow of the recipient and attacking the recipients tissue so these T lymphocytes are usually killed by the recipients body but however maybe the patient has a suppressed immune system and they didn't attack these t lymphocytes whenever they're getting the transfusion and these hea lymphocytes from the donors start attacking that marrow and what can happen is they can have a fever and this really peculiar rash all over the body it'll be on the hands and the feet as well with GI issues diarrhea nausea inflammation of the liver and you want to talk you want to tell your patient you know if you start having this rash from head the this fever diarrhea all this stuff for several weeks after you've had this blood transfusion you want to report that to your doctor other complications that can arise that really aren't immune related it's like septicemia where the blood is contaminated so as a nurse again it's really important that you start that transfusion promptly after receiving it from the blood bank or the blood is contaminated with a disease now this isn't as common because we have strict screening guidelines but there can be a risk for hepatitis B C or HIV etc also that circuit or e overload we are talking about or developing high iron levels and this happens with people who've had frequent blood transfusions now let's talk about if your patient does have a blood transfusion reaction and I have seen this this does happen so keep it in the back of your mind so the first thing what you want to do is stop the transfusion and you want to note mentally what time this occurred what time you stopped it because you'll be documenting this later on also you're going to disconnect the blood tubing at the access side and replace it with new tubing and have some point nine percent normal saline running to keep the vein open then you're going to notify the doctor and the blood bank of what's going on but during all this you're going to be staying with the patient at their bedside you need be watching them you need your eyes on them so this is a great time to call in other people on the floor to be helping you make these phone calls to be doing these things next you're going to be monitoring those vital signs every five minutes looking at them watching them looking at that respiratory status it's not compromised or they have an allergic response what's going on now whenever you contact the physician depending on what type of reaction they suspect the patient's having or how severe it is what's going on they may order some medication so it varies some things they can orders like corticosteroids which is going to suppress that immune response along with fluids helping flush out that free hemoglobin that's in the body getting it out we want out of the body anti hit anti histamines to decrease that immune response anti-pirate ik stu decrease that temperature vasopressors this can help if there have an allergic response like epinephrine to open up the airways there's a lot of time those Airways clamp down and they can't breathe or like dopamine to increase renal blood flow or diuretics also some labs are going to be ordered they want to look at those claudine levels because remember if this is hemolytic type because a lot of times they don't know what type of reaction this patient is having they want to look at those clotting factors because they're at risk for di C where all their clotting factors are just going to be depleted and they can bleed out and die looking at those electrolytes looking at the renal function how's our kidneys and other blood levels in addition you'll be collecting urine urine on them looking for the free hemoglobin that's came from those red blood cells I have lysis and whenever you are disconnecting your tubing over here do not throw it away don't throw any of it away because you'll be sending that one with the leftover blood and any other documentation to the blood bank who's going to test it look out and see what went wrong and of course you're going to document you want to document the time it happened what actions you took what the patient was given if you gave them anything what labs you drew all that and how the patient is currently doing okay so that wraps up this review over blood transfusion thank you so much for watching don't forget to take the free quiz and to subscribe to our channel for more videos
Nursing_Skills_Videos	Achilles_Heel_Deep_Tendon_Reflex_Test_Nursing_Head_to_Toe_Assessment.txt	hey everyone it's Sarah with register nurse ari.com and in this video I'm going to show you how to check the Achilles deep tendon reflex so let me show you how it's done okay now let's check the Achilles Achilles is going to look for L5 to S2 and this is best done having the patient dangle their feet just like whenever you're checking the patellar you usually want to check those two together and um what you're going to do is you're going to dors dorsiflex the foot up like that and you will see the tendon which is located right above the hill um it's right there and what we're going to do is we're looking for the foot to plant our Flex so it's going to go down like that to list a response but in order to do it you're going to dorsy flex the foot by supporting your hand underneath it and just tapping the bottom of it and flex and we'll do it again okay so that is how you check that deep tendon reflex now check out my other videos on how to check the other four deep tendon reflexes and consider subscribing to this YouTube channel
Nursing_Skills_Videos	Eye_Assessment_Nursing_How_to_Assess_Eyes_for_HeadtoToe_Assessment.txt	this is cereth registered nurse re and calm and in this video I want to demonstrate how to assess the eyes now if you would like to watch a complete head-to-toe nursing assessment you can access here in the card or in the YouTube description below a link to the video that will show you how to do that now for this skill what you want to do is you'll want to provide privacy to the patient wash your hands and tell them what you were doing and equipment you will need for this is a pin light so let's get started we're going to inspect the eyes first and we're looking at several things we're looking at the eyelid we're looking at the sclera which is the white of the eyes we're looking at the iris we're looking at the pupil and we're looking at the conjunctiva so you shouldn't see any swelling of the eyelids you should see that the sclera is Y and shiny it shouldn't be yellow like in jaundice and the conjunctiva when you pull down the lower lid have the patient look up it should be nice and pink it shouldn't be red you shouldn't see any drainage or anything like that and look at the eyes how do they set in the eye socket is are they equal for instance is there any strabismus is there a cross eye where one eye turns in more turns out or up or down and these eyes are normal there's no strabismus next you want to look at anisocoria where you have where one pupil would be smaller than the other people are they equal in size normal pupils should be 3 to 5 millimeters in their measurement and here his are about a 3 and they are equal next what we're gonna do is we're going to assess some cranial nerves we're gonna be looking at cranial nerve three which is ocular motor for troq Euler and then 6 which is abducens and we're gonna do several tests to check their function the first one what we're gonna do is we're gonna be looking for any involuntary shaking of the eye called nystagmus and how we're gonna do that was we're gonna take our pin line we're gonna hold it about 12 to 14 inches away from the patient's nose and then what I want you to do is keep your head still don't move your head and just use your eyes to watch where I move the pin line and as you're doing this you're going to do you're going to perform it in the six cardinal fields of gaze and you're just going to move it and you're looking for any involuntary shaking of the eyes so here we go next we're going to see how reactive the pupils are to lie and to do that we're going to dim the lights a little bit and we're gonna have the patients stare off at a distant object that helps dilate those pupils and then we're going to shine using our pin line in at the side and we're gonna see how that pupil response is she constrict and then on the other side it should constrict as well so say their baseline pupil size was like three millimeters it should go down to one milliliter and it should happen on both sides okay so being stare off at that object rod on the wall over there for me okay and that dilates the pupils and we're just going to shine light in at this side okay constrict constrict okay I'm dilate again and then go over to the other side do the same again and they both constricted and equal size next what we're gonna do is we're gonna check for accomodation and how we do that is we turn the lights back on we just previously had them dimmed but we now make it light again we're gonna have him stare off at a distant object that helps dilate the pupils and we're going to take a pin light you can use a pin light finger and you're just going to slowly move it inward to the nose and what you're looking for is that those pupils constrict they accommodate and the eyes cross while looking at the pin line so here we go stare off in the distance please and I don't want you to move your head or anything just keep it real still and just follow this pin light okay ready okay so now we can document because we just checked all of the things with the eyes we can document that the pupils are equal round reactive to lie in the accommodate so that's where that acronym perrla comes into play okay so that wraps up how to assess the eyes and don't forget to check out that video on how to perform a complete head-to-toe nursing assessment thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	Costovertebral_Angle_Tenderness_Exam_CVA_Percussion_Assessment_Test.txt	hey everyone is fair thread sterner sorry and.com and in this video I want to show you how to check for cost over to Beryl tenderness also known as CVA tenderness this is found in patients who may have a kidney infection because what has happened is that the kidney becomes inflamed from the infection and then when you take your hand put on the bat and thump or percuss it will cause the patient tenderness so let me show you how to do that okay so what you want to do is you want to find the last rib posts király and that is the 12th rib so if you fill in between the ribs be about right here and then you're going to find the spine and underneath the 12th rib and between the spine is that cost over to Beryl angle and about right here is where your left kidney is located and over here is where your right kidney is located so what we're going to do is we're going to take our hand and we're just going to put it over here we're going to make a fist and we're going to thump our hand and we're going to ask the patient if those have any pain okay do you have any pain with that okay and then we're going to do it to the other side you have any pain okay and if they had pain that would indicate a possible kidney infection thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	Carotid_Pulse_Point_Examination_Palpation_Location_Nursing_Skill.txt	hey everyone it's sarah thredson err sorry and calm and today i'm going to demonstrate how to find the carotid pulse point and whenever you're assessing the pulse you will be looking at a few things one thing will be the rate how fast is it along with the strength and you'll be grading it on a scale zero to three with zero being absent one plus it's week two plus it's normal and three plus it's bounding and then you'll want to look at the rhythm is it regular or irregular and this artery is most commonly assessed during CPR in an adult and it supplies our brain and our head with blood now whenever you are assessing the carotid artery you will assess each side at a time you will not do it bilaterally because we don't want to stimulate the vagus nerve which can drop the heart rate and decrease circulation to the brain to find this pulse point we'll use the landmarks of the jaw because we're going to go below the jaw we're gonna have the patient tilt the head like this and we're gonna find the trachea and the sternocleidomastoid on't forget to check out our other videos on how to check the other pulse points on the body thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	Hand_Hygiene_for_Healthcare_Workers_Hand_Washing_Soap_and_Water_Technique_Nursing_Skill.txt	hey everyone at CR thread sterner Sarang calm and in this video I want to demonstrate how to perform hand hygiene by using soap and water there are two ways a nurse can perform hand hygiene one way is through using soap and water and another way is through using an alcohol-based hand rub so when do you want to perform hand hygiene you would always want to perform hand hygiene before and after patient care after coming into contact was like a body fluid or an open womb or touching something close to the patient like a bedside table the hand railing on the bed or after removing your gloves before eating and after using the bathroom now when would you use soap and water versus the alcohol-based hand rubs well according to CDC Gove's guidelines you would use soap and water when your hands are visibly dirty after known or suspected exposure to Clostridium difficile which is c-diff if your facility is experiencing an outbreak or higher endemic rates after known or suspected exposure to patients with infectious diarrhea during norovirus outbreaks if exposure to bacillus anthracis is suspected or proven which is anthrax before eating and after using a restroom so now I'm going to demonstrate how to perform hand hygiene using soap and water so first you need your supplies of course you need soap you need some paper towels and you need running water now faucets vary depending on where you work some are automate oh you just have to swipe your hand underneath the water will come on or you use a petal or it has little faucet handles that you have to use to turn on and off now generally it's best not to wear jewelry during patient care because that jewelry can Harbor germs in some areas in the hospital like surgery for instance actually prohibits a person from wearing jewelry all together but usually on some units you can wear like a simple wedding band so if you do wear a wedding band and you're wearing it during patient care you need to keep that wedding band on whenever you're performing hand hygiene because you want to clean it because underneath that ring it can Harbor germs as well so first what we're going to do is we're going to turn on our water and we want our water to be warm not too hot because if it's too hot that can dry out the skin and that can cause you to get cracks in your skin be really uncomfortable for you so make sure it's warm and you want to be careful not to let your scrubs or anything like that touch the inside of the sink because it's very dirty and you'll become contaminated so once your water is warm you want to wet your wrist and your hands and be sure you have your hands lower than your elbows because we don't want the germs that are already on our dirty hands to travel up our arms after you've wet your hands you want to put the soap on your hands and you're going to put about 1 teaspoon of soap which equals about 5 milliliters and a lot of soap dispensers are automative and they give you the amount of soap you need so we're gonna take that soap and we're going to lather our hands and our wrists with the soap now we want to scrub and what we're gonna do is we're going to scrub this up on our hands using circular motions because this is going to help remove those germs and the things that are sticking to our hands and we want to cover all the areas of our hands especially those small crevices because that's where bacteria likes the high we're gonna do this for 20 seconds so first we're going to go and scrub our palms of our hands notice I'm doing a circular motion then we're going to get the back of our hands and do both using those circular motions then we're going to do our thumbs and we're going to do each finger individually making sure we are getting around the fingernail area as well and then you want to get those knuckles so really rub the knuckles up against your opposite hand it's also going to get the outside of that finger now as well then we want to get in between our fingers because again germs love to hide in little crevices that are hard to get to so we make sure we get that then we're going to pay attention to our fingernails and we're gonna take our hand we're gonna go to the opposite hand and we're gonna just get underneath the fingernail right in there just scrape around remove any germs that could be underneath those fingernails then what we're going to do is we're going to get our wrists and go about one inch above the wrists as well now we are ready to rinse the soap off so we're just going to put our hands underneath the water and let the water go downward we don't want it to go upward rinsing all of the soap off from the wrists down to the fingertips once you get them rinsed you want to take your paper towel and you're just going to Pat your hands dry you don't want to scrub your hands with this paper towel because that can damage that top layer of skin and we're going to discard this wet paper towel then we're going to grab a dry paper towel and turn off our faucet and being careful not to touch the faucet with our clean hands and then we're going to discard this as well now we wouldn't have used our wet paper towel to turn off the faucet because the germs that are on the outside of the faucet could have transferred through that wet paper towel and got onto our hands and undone the job of what we just done with cleaning our hands okay so that wraps up this video on how to perform hand hygiene using soap and water thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	Colostomy_Irrigation_Procedure_Care_Ostomy_Enema_Nursing_Stoma_Cleaning.txt	hey everyone it's s with register nurse rn.com and in this video I'm going to demonstrate for you how to irrigate a colostate so what is the purpose of irrigating a colostomy it removes stoil from the colon using an enema which is usually warm water and it can be anywhere from the amount of 500 cc's to a th000 depending on what the physician orders and it's an enema given through the stom so think of it as an enema that you would give through the rectum it's sort of doing the same thing but we're using the stom and this is usually a physician's order you have to have an order to do this and um they may order this if the patient isn't putting out a lot of stool through their stom maybe they're constipated or the patient has chose to do bowel training and um patients what they can do whenever they're doing bowel training they can train their colon to put out at a certain time during those irrigations so they'll be free free from wearing a continuous pouch throughout the day which we be putting out stool throughout the day so it gives them some free time now what are some key points to remember about this okay patients who are going to have the best success with irrigating their colostomy are patients with descending and sigmoid colostomies why why are they going to have the best success well let's look back at the anatomy we have over here a colon and we have the descending and the sigmoid now whenever the patient has one of these Aames they have majority of their large intestine still there and what is the role of the large intestine remember from our last video over colies and ilos the role of the colon is to absorb water so as the stool travels through here the water is getting absorbed so once it hits this area the stool is going to be more formed so it's going to be easier to irrigate and get our bowel trained so um you may see the osy over here where which would be a descending or you may see it over here which would would be a sigmoid now patients who have an IL ostomy are not candidates for this because the stool that's coming out it's connected over here to the ilien these patients have liquid stool it's never formed and it's very rich in electrolytes and fluids so we don't want to be messing with that area so they're not a candidate for this now whenever you're doing bowel training the patient needs to know that it's going to take about 6 to 8 weeks to get the colon trained so they need to know that it takes a little bit of time and a lot of dedication but can have good result for them if they choose to do this not all patients choose to do this um the result will be like I said earlier and they will not have any bowel movements in between that irrigation and it frees them up from wearing that pouch instead they will wear a stomach cap over the stom which is a very small Clos pouch system and they'll perform the irrigations once they get trained maybe every day or every other day depending on the patient's bow habits now some other things whenever you are training the bowel you'll want to teach the patient and perform this at the same time every day why because we want the colon to get used to putting out its contents at a certain time every day so a patient can pick what time works great for them to do this and get the colon used to it and you're going to perform it best time is about 1 hour after a meal because this is when the colon is the fullest you're going to get the most success of helping it get empty um as a nurse what you want to remember and let the patient know this because this process takes time from start to finish it takes about 1 hour so about 5 to 10 minutes for actually instilling the fluid through the stomach and then another 30 to 45 minutes for the colon to end empty itself um generally it comes out in episodes the stool will and for it takes about 30 45 minutes for it to be completely emptied so keep that in mind and this is normally either performed in the bathroom with the commode or in the bed with the bed pan whatever you choose to do and whatever the patient wants to do now let me go over the supplies that you will need to perform this you will need an irrigation bag that holds at least 1,000 cc's of water you will be putting in warm warm water in this bag generally you don't give any more than 1,000 CC's it's usually between 500 to 1,000 also um it will come with some tubing and you may have to connect the tubing to the cone because this cone will be used to insert into the stom when you give the enema and you'll just connect it right here and then here is your little regulator that you will turn on or off as you're letting the water instill into the stom then you will need an irrigation sleeve this is where the water in the stool is going to collect um once it starts draining out and um a lot of times these sleeves come with a clip and this clip will go at the end of the sleeve just to close it shut so the contents doesn't spill out while you're doing this sometimes people just cut this bag to the certain length of what they need and put it inside the commode or the bed pan and just allow the contents to flow through that but for this demonstration we're going to be using the clip and we'll be emptying it that way and I wanted to show you this this is a flange that goes over the stomach it's part of the two-piece system and um what you will do if a patient's having the two-piece system you will just simply take your irrigation sleeve and just clip it snap it onto the barrier so whenever you are doing this procedure it'll allow the stom to let the contents collect in the sleeve but if you're using a one piece system you'll need to get a face plate and an ostomy belt to help you do that so make sure that you get the right supplies before you start this procedure also another thing you'll need is water soluble lubricant because you will be using this to lubricate the cone before you insert it into the stom you will need several pairs of gloves because you'll be changing them in between and also it's a good idea to wear a gown to protect your skin and your scrubs from getting um splashed with stool and the contents coming out especially if you're going to just be letting it drain down into the comode or the bed pan because there's a risk of splashing and you'll want some washcloths with warm water to provide Stomach Care before and after the procedure and some towels just to protect the patient's gown first perform hand hygiene then put on your gown and gloves now fill your irrigation bag up with warm water and before you do that the trick make sure your regulator is in the off position because you don't want to fill it up and all your water spills out after you fill the bag you will want to hang it and hang the bag where the bottom of the bag is about shoulder length to the patient because you want it at this level so the solution doesn't go in too fast which can cause some discomfort and cramping to the patient now so we have water in our bag but the most important thing you want to do is you want to Prime this tubing just like you would an IV tubing because we don't want all this air going into the patient gut it's going to cause them some major discomfort so what you want to do is you just want to um open the valve up a little bit on your regulator cuz remember we closed it when we filled it and we're just going to fill this whole tubing with water so I'm going to open it and it's going through there we're making sure we get all the air bubbles out and there we go okay so our tubing is nice and primed and we're ready to go now what you want to do is you want to clean the stom you want to remove any stool that may be on it so um you'll just take a washcloth with warm water and just gently clean the area make sure it's nice and clean then dispose of that and then what we want to do we want to lubricate our smallest finger and um what we're going to do is we're going to be inserting it into the stom because we want to feel the angle of how the bowel is because that's how we're going to insert our cone so just put a little lubricant on the finger and um whenever you insert into the stom be aware that it can tense up or freeze up a little bit and if that happens just stop and then um Let It relax and then gently fill Don't Force so I'm just going to gently insert and just fill the angle okay now what we're going to do is we're going to change our gloves perform hand hygiene and put on a new pair of gloves now we're going to attach our irrigation sleeve onto the face plate of the flange now um before you do this if you're not using the commode or a bed pan Which um would allow the the stool to go into that instead you're going to clip the bag don't forget to clip your bag so take the clip and then just take it towards the end of the bag and just snap it shut and then take your irrigation sleeve and if you're using the two-piece system you'll just attach to this if you're using the one piece you'll use the face plate and the omey belt and it'll just snap on there and stay in place okay now we're going to lubricate the comb but first what you want to do is you want to open that bag of the irrigation sleeve a little bit first cuz you're going to get lubric on your hands and when GL gloves and lubricant do not go together it's really slippery so we've opened the sleeve of the irrigation sleeve and then we're going to lubricate the cone and get it nice and lubricated so it's easier to insert into the stom and it's not painful for the patient so after you get it lubricated you're going to go in and remember what angle that the stom was at so that will be the angle that you insert the tone and you'll insert it just enough without um causing too much force but you want it in enough where you when you instill the contents it doesn't leak out now you're going to open up the valve of the irrigation bag by turning it to on and you're going to do it slowly so um the water goes in slowly you don't want it too fast or too slow if the patient starts reporting cramping or nausea turn it back off let them feel better have them take a breath and then start slowly again and remember this process takes about 5 to 10 minutes and we wanted all of this to infuse of what the doctor ordered after you've instilled all of the water wait about 10 seconds then remove the cone from the stom and dispose of that properly and then be sure to seal the sleeve of the irrigation sleeve and um have the patient hold the contents for about 30 to 45 minutes and you will notice that over that time spurts of stool and irrigation fluid that you instilled will come seeping out and it'll look something like this and as you can see here the irrigant in the stool has come out and um what we're going to do is we're going to remove this irrigation sleep leave and we're going to provide Stomach Care replace the flange the Skin Barrier and you'll want to chart this per your hospital protocol however they require and you'll want to talk about what you done how you explained it to the patient how they tolerated it and how much um came out whenever you measured it and your documentation could look something similar like this if you were using the fdar format okay so this w up how to irrigate a colostomy thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	Vital_Signs_Nursing_Respiratory_Rate_Pulse_Blood_Pressure_Temperature_Pain_Oxygen.txt	hey everyone it's sarah thread sterner sorry and calm and today I want to demonstrate how to take bottle signs first what you want to do is you want to gather your supplies typically what you're going to need is a stethoscope with a blood-pressure cuff along with a thermometer of some type a pulse ox a watch and some gloves along with a disinfectant wipe to clean the items that are not disposable then what you want to do is you want to perform hand hygiene and Don the appropriate PPE if necessary like if your patience and contact precautions you'll want to put on the correct PPE now what is collected during the bottle sign measurement well you'll be checking the patient's blood pressure heart rate respirations and temperature in addition you'll be asking the patient their pain rating which is sometimes referred to as the fifth bottle son along with the collecting their oxygen saturation so I've arrived to the patient's exam room and I've performed hand hygiene now what I want to do is I want to introduce myself to the patient and tell them what we're going to be doing so hello my name is Sarah I'm a nurse here and I want to be taking your vital signs is that okay with you yes okay then you want to do your patient identifiers by looking at their armband having them tell you their name and their date of birth then I like to start with the easiest thing which is pain and so I'm going to ask him his pain level now this is a very easy and important assessment tool because high pain ratings if the patient is in pain it can alter their vital signs it can increase their heart rate their blood pressure and respirations and it's really important especially to ask a patient their pain level if they've just had surgery or some type of trauma so to assess pain levels you can do that with various skills most commonly we use the 0 to 10 numerical scale so can you tell me your pain with zero being no pain at all to 10 being the worst pain you've ever had what what's your pain rating zero okay he says he's having no pain that's easy but let's say that they he said that his pain rating was an 8 well you would want to ask some more questions you and I say where's your pain located at and please can you describe it for me like burning as a radiating things like that and then you want to document that the numerical rating along with the words that the patient used to describe the pain in its location next we're going to measure the patient's oxygenation status and to do that you can use a portable probe like this one or one that connects to a bedside monitor and to do that you're going to place the device on the nail bed because that's where it's going to obtain the reading so make sure that you pick some fingers that have good circulation they're nice and warm in pink so we'll turn on our device and we will place it on the finger and let it get a reading and here the patient's oxygen saturation is 97% a normal oxygen saturation is anywhere between 95 to a hundred percent and below you can also see the heart rate as well but here in a moment we will actually check the heart rate and then you'll just want to remove the device and if it's like a portable one like this you'll want to clean it with a disinfectant wipe and then document your findings now we're going to collect the patient's body temperature and some things you want to remember about body temperature is that in an adult it can vary it can be anywhere between 97 to 99 degrees Fahrenheit with the average being about 98.6 degrees Fahrenheit orally and an adult it's considered a temperature if the temperature is greater than a hundred point four degrees Fahrenheit now the temperature reading will depend on the route that you use and you can take a patient's temperature various ways like orally the forehead via the temporal artery tympanic lis which is via the ear rectally or axillary via the armpits and a rule of thumb to remember is that rectal and tympanic temperatures will be one degree higher than the oral route and temperatures that are collected via the axillary or the temporal route will be one degree lower than oral temperatures so we're going to take the patient's temperature using the temporal artery and we're going to use this device so what you want to do first is you want to use a probe cover if your device has one that just protects it from becoming contaminated and what we're gonna do is we're going to hold the probe flush up against the skin at the center of the forehead we're going to take it and scan it across the forehead to the hairline and look at her reading and before we do that you want to make sure that the forehead is clear of any type of hair or anything because this probe needs to be making contact with the skin if anything comes into contact with it can throw off the reading so we're going to put it flush against the skin and hold the button in on the device and you'll hear it beeping and scan it to the hairline and look at our temperature now if your patient was sweating on the forehead because a lot of times whenever patients have fevers they can sweat you would want to do it the same way probe up against the forehead in the middle hold the button down scan across the forehead to the hairline but you're also gonna go behind the ear because sweating will decrease the temperature and it's very vascular back here behind the ear and that will just help us obtain a proper reading then what you're gonna do is you're going to clean your device and document your finding and if you didn't take it orally you want to make sure you document the route that you actually took the temperature next we're gonna check the patient's pulse and as we feel the pulse we're going to be looking at several things of course we're going to be counting the rate but we're also going to be feeling the strength of the pulse and we will be grading it on a zero to three plus skill with zero being the pulses absent one plus week 2 plus normal and three plus bounding and the rhythm is the pulse regular or is it irregular now in adults the most common site to use to check the pulse is the radial artery because it's really easy to access so it's found what you want to do is find the thumb and it's found below it in this wrist area along the radial bone hence why we call it the radial artery and whenever you're checking the pulse have the patient they can set and bad they can lie down and you'll want to support their arm extended out in some horde it and you're going to use your first three fingers to feel the pulse don't use your thumb your thumb actually has a pulse in it so use your first three fingers and find it within that area I just told you and lightly just touch it don't press too hard and feel the bounding of the pulse and what you want to do is you want to count it for thirty seconds if the pulse is regular and multiply that number by two if it's irregular count it for one full minute so his heart rate is 82 its regular and it's two plus and a normal heart rate an adult is 60 to 100 beats per minute now what we're going to do is we're going to keep our fingers here because what we want to do next is check the patient's respirations and if you tell a patient that you're checking the respirations they're going to alter the way that they're breathing so we're gonna stay in this same position and assess respirations and when we're assessing respirations we're looking at a couple things first of all the rate a normal breathing rate in an adult is 12 to 20 breaths per minute we're also looking at the depth is it labored or unlaid and the rhythm are the breaths regular or irregular and I have found the easiest way to do this is really look at the patient from the side and watch their game their clothes are they rising and falling because one rise and one fall equals one respiration you could also sometimes just gently take your hand put it on their back and fill the rise and the fall of the chest and so you will count that for 30 seconds if their breathing rate is regular and then multiply that by two but if it was irregular you would need to count for one full minute and then document your findings and lastly what we're going to do is we're going to measure the blood pressure and to do that we want to make sure a patient is sitting down with their arm at heart level and their legs are uncross now they're lying in bed you would want to make sure that this arm is at heart level then what we're going to do is we are going to get our stethoscope our blood pressure cuff and you want to make sure you get the right size cuff for your patients arm because if you use too big of a cuff or too small of a cup it can throw off the reading and what we're going to do is we're going to palpate the brachial artery because this is the artery we're going to be listening to to get our blood pressure because we're going to be getting our systolic number which is that top number and this is the first sound we hear and then our diastolic number which is the bottom number and this is the point where we no longer hear the sound so whenever we're looking at the gauge of our blood pressure cuff we want to make sure we're really noting those points because it's going to tell us our systolic and diastolic number so what we're going to do is we're going to put our cuff on our patient and we want to make sure we find the brachial artery this is the artery we palpate that we'll be using to determine our blood pressure and it's found in the bend of the arm so we're going to find it and it is located here and we're going to look on our Kufner cuff has these arrows and because this is the left arm we're going to make sure that this arrow is pointing in that direction of where that artery is so you're gonna put the cuff up about two inches above the bend of the arm first what we want to do is we want to estimate the systolic pressure so we want to find that number to do that we're going to palpate the brachial artery and we're going to inflate the cuff until I no longer feel the brachial artery and at that point when I no longer feel it I need to make sure I'm looking at this gauge to know that number because that number is our estimated systolic pressure number then when I go to take the blood pressure I'm going to inflate the cuff 30 millimeters of mercury more than that estimated number now the whole reason for doing that is because we want to avoid missing the oscillatory gap that can occur in some patients all patients have it but some and it's usually patients with hypertension because the oscar tory gap is like this abnormal silence that occur and it will throw off whenever you actually hear that first sound which is your systolic number so I'm inflating the cuff by filling on the artery and I'm going to note the point where I no longer feel the artery which is about at the hundred then I'm going to deflate it completely and wait about thirty to sixty seconds and then we'll take the blood pressure so we're estimated systolic number is a hundred now I'm going to inflate the cuff to a hundred and thirty and that will avoid missing the oscillatory gap if one was present so I'm going to take my stethoscope put it in my ears you can use the bell or the diaphragm of your stethoscope I like to use the Bell because it's best at picking up low-pitched noises so we're going to place that over the brachial artery do it lightly don't fully compress it because you can include the artery then we're going to inflate our cuff to a hundred and thirty millimeters of mercury and we're going to let it fall about two millimeters of mercury per second and we're listening for that first sandwiches our systolic number okay is 104 and we're listening for that last sound and it was 78 so the blood pressure is 104 over 78 then once you have your reading make sure you fully deflate the cuff full of air and you're going to take the cuff off of your patient of course and clean it if it's not disposable and you will document the blood pressure and what arm you took it in now water normal blood pressure readings according to the American College of Cardiology 2017 updated guidelines a normal blood pressure is a systolic less than 120 and a diastolic less than 80 elevated blood pressure would be considered a systolic of 120 to 129 and a diastolic less than 80 hypertension stage 1 would be a systolic of thirty to 139 or a diastolic eighty to eighty nine and hypertension Stage two would be a systolic greater than 140 and a diastolic greater than ninety okay so that wraps up this demonstration on how to check vital signs thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	Health_Assessment_Nursing_NCLEX_Practice_Question_on_Abdomen_and_Lymph_Nodes.txt	this is cereth registered nurse re and calm and in this video I want to be going over to in click style questions that will test your knowledge on health assessment and don't forget to check out the other questions in this series so let's get started our first question says you're performing a head-to-toe assessment on a patient admitted with abdominal pain during inspection of the abdomen you know the abdominal contour to be round and distended with no masses or lesions present the patient reports that their last bowel movement was one hour ago in the stool was loose in addition the patient states that the abdominal pain is located below the umbilicus and is sharp in quality after inspection of the abdomen you will a perform light palpation on the abdomen followed by deep palpation be percuss the abdomen see auscultate for bowel sounds by starting in the right lower quadrant or D palpate for brewery's and rebound tenderness so in this scenario we are performing our head-to-toe assessment on a patient who is admitted with abdominal pain so we really want to focus on that abdomen and according to the scenario we've already inspected the abdomen we found that the abdominal contour is rounded their stomach is distended but there's no masses or lesions and the patient told us their last bowel movement and they are having abdominal pain so what are we gonna do next since we've already inspected well what you want to be doing is you want to be pulling from that health assessment knowledge and remembering that particular order of how we execute our head-to-toe assessment normally first we inspect then we palpate percuss and then lastly we auscultate now with the abdomen it's different with every other system you're gonna do that except with the abdomen with abdomen we're gonna change it up a little bit first what we're gonna do is we're gonna inspect which is what you do first up there then second we're going to auscultate then thirdly we're gonna percuss and then palpate so why we doing this second instead of last okay the whole reason we change it up is because we want to prevent all train those vowel sounds because think about it whenever you go and you do light palpation and deep palpation on the abdomen you're manipulating and messing with those intestines squishing that stuff around so that can produce a bowel sounds so initially we just want to look at that stomach and then we want to listen with our set the gut stethoscope to those sounds see if we hear anything so that's the whole reasoning for it so whatever you're looking at those options you are looking for an option that is talking about performing auscultation listening with your stethoscope and after reading all of our options the only one that talks about listening with a stethoscope is C so we're answer is C we're going to auscultate for bowel sounds by starting in that right lower quadrant then after that we would listen for those vascular sounds on the abdomen of the aorta the renal artery the iliac artery all those and we're listening for brewery's then we would perk us and then lastly we would palpate so that's why a is wrong that's why B is wrong it's talking about percussion and then D is wrong for a couple reasons he says pal VapoRub Ruiz no we don't have a four breweries we listen for breweries that swishing noise that shouldn't be there so that's why that's wrong and of course it's talking about palpating for rebound children tenderness and we would do that last so our answer is C now let's look at our next question you're performing a head-to-toe assessment on a patient while palpating the lymph nodes of the neck the patient reports tenderness at the following location when you document the findings of the head-to-toe assessment you will note that the patient felt tenderness at which lint node site a preauricular B submandibular C superficial cervical or D jugular digastric so to answer this question you have to know where all those lip node sites are in the neck and what they are called so let's review those so if you start from the top and work your way down you will first you will find in front of the ears the preauricular then behind the ear the post our Rick Euler and a little bit down from that at the base of the head the abscissa tool then a little bit down from the ear in the neck you will find the jugular digastric then down from there you'll have the sub mandibular and this is where below the mandible and then a little bit under the chin you can find the submental and up a little bit next to where the jugular digastric is you will find the parotid lymph node and as you work your way down the neck you will find the superficial cervical and then the ones down in the yellow is the deep cervical chain and these go down through the anterior and posterior sides of the sterno mastoid muscle so you will find those in those different areas and then a little bit posterior to that is the posterior cervical and right above where the clavicle is you will find the supra clavicular so the answer to this question would be jugular digastric that is where we would document that the patient felt the pain okay so that wraps up this review of those two health assessment questions and don't forget to check out the other questions in this series thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	Babinski_Reflex_Plantar_Reflex_Test_Nursing_Head_to_Toe_Assessment.txt	hey everyone it's Sarah with register nurse rn.com and in this video I'm going to be going over the planter reflex what I'm going to do for you is I'm going to go over the basics about this reflex and then I'm going to demonstrate for you on a person what a positive and negative planter reflex looks like with the babinsky sign so first let's talk about what the planter reflex is this is a superficial reflex found on the bottom of the foot now this is different then the Deep tendent reflexes that we covered in the previous videos in that these reflexes are due to the receptors found on the skin and what we do is we elicit a reflex through the skin rather than the muscle like how you doing the Deep tendons now whenever you do a planter reflex you're looking for the babinsky sign and the babinsky sign which you'll see here in a minute you'll see a positive and a negative one and whenever you see a positive babinsky sign this is this is is abnormal you will see dors dorsy flexion of the big toes and Fanning of the toes so in other words you're going to see the big toe and the toes dorsiflex which means they're going to move out like that and this is common now this is a normal sign if you are doing this on an infant less than 24 months of age but if you're not it can be indicative of um upper motor neuron disease now if a negative binsky sign which is normal you want to see this in someone um 24 months or older it's where you have planter flexion and curling of the toes with uh flexion of the forefoot so you'll see the toes sort of planter Flex in and curl so now let me demonstrate for you what that looks like okay for the planter reflex we're taking L4 to S2 and what we're going to do is we're going to lightly stroke up the lateral side of the sole of the foot inward through to the ball of the foot to make like an upside down jet and what we're watching for we're going to look at the toes and we want for a normal response for it to um plant our flex and for the toes to curl inward positive binsky sign would be whenever the foot dorsiflex and the big toe fans out so let's see what we get okay and that was planter flection notice that the toes they curled inward let's do it again and they curled inwards so this is um normal okay let's see what a positive babinsky sign looks like and notice we had dors flexion of the foot and the toes faned out let's do it again okay so that is how you check the planter reflex be sure to check out my other reflex examination videos and thank you so much for watching and please consider subscribing to this YouTube channel
Nursing_Skills_Videos	MeteredDose_Inhaler_MDI_Demonstration_Without_Spacer_Nursing_Open_Closed_Mouth_Technique.txt	hey everyone its era thread sterner sorry and calm and in this video I'm gonna demonstrate for you how to use a metered dose inhaler using both the open mouth and closed mouth technique so what is a metered dose inhaler this is an inhaler used to deliver a set amount of medication into the patient's lung and the medications include like bronchodilators like albuterol a brand name for that is like ventolin or pro air along with corticosteroids such as fluticasone which is known as flovent and patients who have respiratory diseases like asthma COPD may be prescribed these inhalers now these inhalers can also be used and be ordered as a combination where there's one inhaler but there's two different types of drugs in them like symbicort which would contain a bronchodilator and a corticosteroids so you want to keep that in mind and always look at what your patients going to be taking another thing is is that you want to teach the patient proper technique when using these inhalers because if they don't use them correctly what will happen is that not a lot of the medication will get down into their lungs it'll just stay in their mouth and that is a big issue with some of these medications like corticosteroids so whenever you are teaching the patient how to use an inhaler try to get a spacer this is what it looks like and it attaches onto the inhaler and they inhale through that which I will be demonstrating in the next video how to use but if and then hey if a spacer is not available you can use the open or closed now technique and always go by whatever the physician has prescribed for the patient while the patients will tell you what they've been using and you need to make sure that they're doing it right now why would you want to use a spacer well the spacer is going to be easier to use than that open mouth the closed mouth technique because whenever you're using this inhaler with those techniques you have to simultaneously press the inhaler down and inhale at the same time which takes a lot of practice so if you have a really young patient or a patient who's older who has issues with holding the inhaler that can be troublesome also then hailer in the chamber is going to allow this in a sense lead time for them to inhale the inhaler substance so they don't have to press it down simultaneously and it's going to decrease the amount of medication that's going to be left in the mouth which with corticosteroids we don't want that because it can cause mouth irritation like thrush so what we're going to do first is we're going to perform hand hygiene and we're going to perform the patient's five rites we're going to make sure that we have the right patient the right drug the right dose we're doing that at the right time and the right rap and you always want to look at what you're giving because whenever you're doing these inhalers or some things you want to keep in mind okay say you were going to be giving a bronchodilator inhaler and a corticosteroid inhaler so you have two inhalers it's very important you know which one you need to give first the first one you want to give is the bronchodilator because it's going to open up those lungs and die like that then you need to wait five minutes and then administer the corticosteroid because then the corticosteroid can get in there and do it straw by decreasing the inflammation now say you were giving two puffs of a bronchodilator and that was it or two puffs of the corticosteroids and that was if you would first do a puff of the bronchodilator and say they need another Ronco dilator puff you would wait one minute and then give the other ones so if you're giving the same medication just two different doses you're gonna wait one minute but if you were going to give a bronchodilator then a corticosteroid you would wait five minutes because we need five minutes for the bronchodilator to do its job and then another thing you need to remember is that after the patient uses then hailer after doing a corticosteroid they need to rinse they need to gargle and rinse their mouth with water and then spit that water out because the corticosteroid whenever he comes into contact with those mucous membranes and the mouth it can cause thrush so it's very important you get your patient to do that next what you want to do is you want to prime then hailer so you're going to prime the inhaler if it's the first time you're ever using that if the patient hasn't use it in a week or more has dropped it or recently cleaned it so before you do that just make sure that your inhaler is in de it's not expired so to do that what you're going to do is you're just going to gently pop this canister off and it will tell you the expiration date this is 2020 so we're in date and the dose that we're using in this inhaler the reason here is a demo dose it's a teaching dose that doesn't like contain medication also never administer medications to a patient if you don't have proper credentials or if you've not been trained and always follow your Hospital protocols so another thing whenever you're going to go priming you want to check expiration date but you also want to make sure that there's enough doses in this inhaler because they only contain so many puffs so a lot of inhalers have a counter on them and it'll tell you how many's last like 2010 however many so look there to make sure you have enough and if it doesn't say it like this one doesn't the box will tell you how many sprays are in each inhaler and this one has 200 sprays so you'll need to pay attention to how many sprays the patient's doing a day and calculate that out for instance say that the patient is doing two puffs twice a day so how many puffs is admin today four and there's only 200 puffs in this box how many days is this inhaler going to last if they do it every day just 50 days so you want to keep track of that now two problem inhaler what we're going to do is we're going to remove the cap and then just hold with your thumb at the bottom and your two fingers at the top and gently give it a shake and depending on the manufacturer is how many sprays you're gonna spray to prime it because each and hailers difference always read the instructions with this one it's four so what we're going to do is just do four sprays now that our inhaler is prime let me demonstrate for you the open mouth technique I'm going to tell you about it and then I will show you how to do it or so what you're gonna do is you're gonna have your patient set up and you're gonna have them hold the inhaler in between their thumb and their two fingers and give it a good shake for about eight to ten seconds just to mix up that medication and really good then they're going to measure out two fingers with between their mouth and then hailer and make sure that inhaler is pointed at their mouth and what they want to do is they want to breathe in and then breathe out through their mouth until they no longer can and then after that simultaneously they want to hit the inhaler down and breathe in slowly through the mouth of the stuff that's coming out of the inhaler and then they'll want to hold their breath for 10 to 12 seconds and breathe out slowly if they need another puff of that same medication they can do it in one minute so let me show you how to do that then after that retype the inhaler and if it was a corticosteroid be sure to gargle and rinse the mouth with water and spit and perform hand hygiene and document now let me demonstrate the closed mouth technique so what you want to do is hold inhaler in between the thumb and the fingers and give it a good shake for about eight to ten seconds to get the medication mix well and you'll have the patient take a breath in and then out through their mouth until they can no longer exhale and then put the mouthpiece of the inhaler in between the tea keeping the tongue flat on the lips sealed around it and they will simultaneously hit then hellar button and breathe in until they no longer can breathe in and then hold it for about ten seconds and then exhale slowly and that will look something like this and after that recap inhaler and again if they were using a corticosteroid have the patient gargle and rinse the mouth and spit with water and perform hand hygiene and document and if they needed to use the same medication again how when would they do it they would do it in a minute but if they were going to be using say this was a bronchodilator and they needed to use the corticosteroid how long would they wait five minutes okay so that wraps up this video on how to use a metered dose inhaler using the open and closed mouth technique thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	How_to_Walk_with_a_Cane_Nursing_Skill_Demonstration.txt	hey everyone this is air thread sterner sorry end calm and in this video I'm going to demonstrate how to walk with a cane before a patient starts using their cane and practicing with it you want to make sure that they are wearing a gait belt for safety in addition you're going to stand on the patient's weak side in case they lose their balance so whenever they start using the cane they want to make sure that they're in the proper position the position that they want to be in they want to make sure that they are positioning the tip of the cane at least four inches from the side of the foot and they want to hold the cane on the strong side of the body so remember that the cane needs to be on the strong side very important concept to remember so how are they going to actually ambulate with this cane well to emulate with the cane what they're gonna do is they're going to move the cane with the weak side together forward so they move the cane along with their weak side together and then they will move the strong side forward thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	SwingTo_Gait_and_SwingThrough_Gait_Crutches_Nursing_Skill_NCLEX.txt	hey everyone in Sarah thread sterner sorry end calm and today we're going to demonstrate how to do the swing to gate and the swing through gate while using crutches and let the names help you so the swing to gate is where the patient will move both crutches forward then they will hint swing or move both legs forward and place them at the placement of where the crutches are located now the swing through is very similar to this they will move both crutches forward then they will move both legs forward can't swing both legs forward but they will swing it pass the crutch placement thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	How_to_Use_Crutches_2_3_4Point_Gait_SwingToThrough_Stairs_Nursing_NCLEX.txt	hey everyone it's sarah thredson are sorry and calm and today we're going to talk about crutches and as always when you get done watching this review you can access the free quiz that will test you on this assistive device so let's get started for exams in nursing school you want to make sure that you are familiar with assistive devices such as crutches canes and walkers and in our next videos we will be talking about canes and walkers so with crutches what you specifically want to know for exams is how they should properly fit the different types of gates you can do with crutches how a patient should go up and down the stairs with crutches and how they should get up or set down in a chair so first let's talk about proper fit what are some things that you need to see in your patient that tells you that those crutches fit them properly well before a patient even starts using crutches the crutches will have to be adjusted based on the patient's height and they can be adjusted at the bottom and at the top and usually on most crutches there's a height range where you will slide the part of the crutch to meet whatever your patient's hype is also there's a hand grip on the crutch that can be adjusted as well so some things you want to remember is that you want to be looking at the gap between the crutch rest pad and axilla the armpit there should be a gap there it should be about two to three fingers width which ends up being about one to one and a half inches and the reason you want this gap is because while the patient is using using the crutch they need to be putting all their weight on the hand grips rather than on this crutch rest pad via maxilla area because in this area you have nerves and if they're putting all their weight on that number one it's really gonna be uncomfortable and hurt but it's also gonna damage those nerves in that area also you want to be looking at the hand grips and where they're lining up on the body the hand grips should be even with the top of the hip line so whenever the patient actually uses the handgrips their elbow will be slightly bent at about 30 degrees now let's talk about the different types of gates while using crutches and you want to make sure that you really truly understand these different types of gates that I'm going to go over because exams love to give you a description of a gate that a patient may be doing while using crutches and you'll have to identify it so before a patient even starts ambulating with crutches at first you want to make sure that they are wearing a gait belt for safety also before they start doing one of these type of gates they're going to start in the tripod position and it looks something like this and it forms like a triangle Hintz looks like a tripod and this is where the crutches are about 6 inches out diagonally from the feet so first let's start out with the point gates we have three of them we have the 2-point gate the 3-point gate and the four point eight now how you can keep these straight is that you need to ask yourself how many points are on the ground hence how many crutches are on the ground and how many feet are on the ground whenever you're looking at the scenario because each crutch counts as a point and each foot counts as a point so the two point we're gonna have two points on the ground at a time whether it's a crutch or a foot so what does it look like well this is where the patient will move the crutch on the injured side so we're gonna say it's the right side so they move the right crutch and they move the left foot together then they will move the left crutch on that non injured side and the right foot together so you have two points next is the four-point game and it's a little bit similar to the two-point gait but each point is moving separately because remember in two-point they were moving together the Rye and the left moving together and then the left and the right moving together but here with four point they're separate so the patient will move the right crutch witches will say the injured side then they'll move the left foot then they'll move the left crutch and then they'll move the right foot and the last point gay is the 3.8 and this is where they move both crutches and the injured leg together at the same time and then they will move the non injured leg and then lastly we have the two swing gates and it's either a swing to gate or a swing through gate and let the names help you so the swing to gate is where the patient will move both crutches forward then they will hint swing or move both legs forward and place them at the placement of where the crutches are located now the swing through is very similar to this they will move both crutches forward then they will move both legs forward can't swing both legs forward but they will swing it pass the crutch placement now let's talk about stairs how does a patient navigate up and down the stairs while using crutches well you want to keep these two straight and what I'm meaning is which leg is going to go first up the step versus which leg is going to go down the step first and to remember that remember good up and bad down so whenever a patient is going up the stairs their good leg is going to go first up on the step followed by the crutches and the bad leg which will proceed and go up the step now whenever they're going down the steps they're going to move the crutches down onto the stab that will help provide stability followed by the bad leg because a bad leg is going to go down and then they're going to move the good leg down on to the step and lastly we're going to wrap up the lecture and talk about how a patient sits down and gets up while using crutches to sit in the chair the patient's going to back up to the chair and fill the chair with the non injured leg and when the patient fills the chair with a non injured leg they're gonna stop and move both crutches over to the injured side for support then the patient's going to grip the hand grips and slightly bend the non injured leg and feel behind them and then set in the chair while keeping the injured leg extended to get up from the chair the patient is going to take the crutches and put them on the injured side for support he's going to keep the injured leg extended and push up on the non injured side and using the hand grips of the crutches then he's going to put the crutches in position thank you so much for watching don't forget to take the free quiz and to subscribe to our channel for more videos
Nursing_Skills_Videos	Fear_of_Needles_Nursing_Tips_for_Patients_with_Needle_Phobia_IV_Tips_and_Tricks.txt	hey everyone cereth register nurse sorry and calm and in this video i want to go over some tips on how to deal with patients who are afraid of needles so as a nurse you are gonna encounter patients who were just really terrified of needles and this can affect both male or female patients and from my personal experience I have found that my male patients tend to struggle with this a little bit more than my female patients for example I had the smell patient who needed his IV changed out his other one had expired so I go in there I get the IV first stick and I'm sitting there flushing it and I wasn't flushing it all of a sudden the patient he just falls back in the bed with his eyes closed and I'm thinking oh no I think he might have passed out and so I look up on the heart monitor he still has a rhythm I feel his pulse still has a pulse I'm trying to keep my IV nice and secure because it's not taped down yet because I'm just flushing it and I'm like sir are you okay and he opens his eyes he's all pale he's like I'm really sorry but I'm just super afraid of needles and I should have let you know and another time I was working at the VA and a clinic drawing blood and a lot of the patients that we would receive have been in war they have killed people they've seen blood and guts and they've been in really really stressful traumatic situations and some of the men before we would actually draw their blood they would request to be laid on a gurney and one man absolutely told me point-blank he's like listen do not even show me that needle before you stick me because he just did not want to see it because it made him pass out so as you can see this is a real issue that some patients experience so as a nurse what can you do to help these patients out first what you want to do is you want to see if your patient does have a phobia of needles and how we can do that is you can look at their body language when you mention that you have to draw blood or you have to start an IV on them because a lot of times patients aren't just going to tell you listen I'm afraid of needles I don't like needles very rarely will they do that because they don't want to come off like they're weak or something like that so whenever you do tell them you have to this procedure look at that body language do they tense up do they look away do they side to the cringe something like that and I can tip you off hey I need to be delicate with this because this patient doesn't like this we don't want to make their phobia worse compared to patients who aren't really afraid of needles you tell them what you're gonna be doing they're like okay when we gonna do it next you'll want to ask your patient you know where would you like me to draw your blood or start your IV because a lot of these patients know where the best place to draw blood or get that vein in that first stick and it usually causes them less pain and they prefer that site so always just ask them if they have a preference some patients don't and some patients do and another thing whenever I've had a patient who has actually told me you know I'm really afraid of the size of the needle what size are you going to use and if in that particular situation I can use a smaller size needle let's say like a 22 or a 23 because they're not gonna be getting blood products or they don't need it for like a special procedure then I'll let them know yeah I'm gonna use the smallest needle I can on you the next size is almost like a pediatric and I have found that that has helped some patients whenever I let them know that it is a smaller needle and whenever you're getting out your supplies and you're setting it up beside the patient before you actually stick them you want to take care and make sure that if the patient is watching you a lot of times patients who are afraid of needles that are gonna look away and not watch you but the ones who do watch you you want to take care that they understand that this huge saline flush is not the needle that's going to be poking them because I've actually had patients look at me and say that's the needle and I'm like oh no this is just a saline flush there's not even a needle on that in that IV packaging with the actual needle in it can look bigger than what it really is because it has the cap over the IV and things like that so you want to make sure that your patient doesn't think that you're just using this huge needle on them then said this is just the packaging over the needle now if you have a patient who is just really struggling with the thought of you you and needle on them to draw blood or start an IV or say this is like the second or third time you have to stick them because there are really hard stick and it's causing them a lot of pain you have options of obtaining an order for some like numbing cream or a numbing agent that will numb my skin there's emilich cream that you put on the side it takes about an hour for it to actually numb the area so keep that in mind or vape oh cool and or something like that now in some facilities that you may have a protocol standing you can do that but in my experience I've only had to do it once and I had to call the doctor to get an order so you always want to keep that in mind now when it actually comes to giving that stick your best tool as a nurse is distraction I have found through all the times I have stuck someone who's afraid of needles is distraction it's my biggest tool and it helps them and it helps me so what I do is as I'm getting my supplies together and I'm cleaning their side I am keeping them in deep a dialogue about something that is interesting to them not about something that's fluffy like the weather or something just random that you really don't think much about you just have this automatic response you want to talk to them about their career some current event that they're really interested in have them talking deep to you because the biggest thing that they're afraid of is that initial stick of that needle they aren't anticipating that pain and they don't want that pain so if you're keeping them distracted they're not going to really know when it happens because you're getting your supplies your cleaning and everything so whenever you go to stick I don't tell patients whenever I'm fixing the stick I just stick and then do my flush and everything like that and a lot of times I have found that as I'm flushing my me whenever I'm done the patient will look at me and say are you already done I'm like yeah and just keeping their mind distracted helped them a lot so I really recommend that so say you do have a patient that passes out like my one patient or actually screams out in pain while you're poking them with the needle or they start crying or something like that you want to validate their feelings and you want to sure that you have your nonverbal body language under control because you don't want to sign or roll your eyes and communicate to them that you think they're just a big baby or something like that because you can make their phobia worse and make them feel inadequate so just always make sure that you let them know that you know I've had other patients that have had issues with this as well so you're not the only person and finally say that you're a new nurse and you just don't have your IV skills and you're drawing blood skills really polished up yet you still need experience which is totally normal because whenever a lot of new nurses start out they're just not good at IVs are drawing blood until they get experience and they do it day in and day out and you have this patient who literally does not like needles and they've told you point Mike you only get one stick well if you're like me when it arrives a new nurse whatever a patient told me that I'm thinking well I'm definitely probably not gonna get it first job because you told me only get one stick and that's just made me more nervous so you probably what you want to do in that type of situation is you want to seek out a more experienced nurse and ask them if they wouldn't care getting this IV for you because the patient is afraid of needles and they only are gonna allow one stick and to help sweeten the deal for another nurse you can tell them hey what do you need done I can do this for you if you can do this for me too like a trade-off so um do that in those type of situations if that does happen but as you get experience and as time goes by you won't want to get someone else to do it for you you'll want to do it yourself because you've got to get experience okay so those are some tips on how to deal with those patients who are afraid of needles thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	Abdominal_Examination_Exam_Nursing_Assessment_Bowel_Vascular_Sounds_Palpation_Inspection.txt	this is cereth registered nurse Arion calm and in this video I'm going to demonstrate how to complete an abdominal assessment and if you would like to watch a complete head-to-toe nursing assessment you can access this card up here in the corner or in the YouTube description below access to that video now before you do this skill you'll want to provide privacy to the patient perform hand hygiene and tell the patient what you will be doing and some equipment that you will need for this is a stethoscope so let's get started now we're going to assess the abdomen and remember we're switching our sequence and how we assess we're going to do inspection auscultation and then percussion or palpation so we're going to auscultation second so whenever you're looking and assessing the abdomen have the patient lay on their back and what we're gonna do is we're going to inspect the abdomen and first we want to ask Ben are you having any stomach issues at all no okay and when was your last bowel movement yesterday morning and how are you urinating do you have any pain while you're peeing do have problems starting a stream any discharge anything like that okay and with your male patients you want to ask about that due to prostate enlargement was starting a stream and if he was female I would ask him when his last menstrual period was and also again ice to be more patient about urinating and things like that now if the patient had a Foley this is the time when you would want to look at the urine inspect the Foley and look at that just conglomerate your urinary system in your GI system together okay so we're inspecting the abdomen we're looking at the abdominal contour and this patients is scaphoid it goes in a little bit you can also have flat round it or protuberant and also we're going to know if there's any pulsations a lot of times in this area right here on thin patients like with being I can see the aortic pulsation in this patients rod above the umbilicus and looking at the belly button and checking for any mass do we see any hernias or anything like that also if your patient had any wounds you wouldn't want to look at that and if they had a peg tube you wouldn't want to assess the site make sure it's not red and ask them how it feels and with your ostomies with your ostomies you want to look at the stoma and make sure it is like a rosy pink color it's not a dusky cyanotic color and it's not prolapsed and look and see what type of stool it's putting out and note that note the smell note when if the bag needs to be changed anything like that so now we're ready to listen to the bowel sounds and what we're going to do is we're going to listen with the diaphragm of our stethoscope and we are going to start in the right lower quadrant and work our way clockwise and we're gonna listen all four quadrants and you should hear five to thirty sounds per minute and if you don't hear any bowel sounds you need to listen for five full minutes and you need to note are these normal are they hyperactive or hypo active so let's listen right lower quadrant we're gonna move out to the right upper quadrant move over to the left upper quadrant and then down to the left lower quadrant ambassy ons are normal now we're gonna listen for vascular sounds and you're gonna do this with the bell of your stethoscope and we're gonna listen at the aortic we're gonna listen at the renal arteries iliac arteries and you could listen at the femoral already arteries if you need it to so you're gonna listen at the aorta artery and it's a little bit below the xiphoid process a little bit above the umbilicus so about right here and we're listening for like a blowing swishing sound that which would represent a bruit okay and none is noted then we're gonna listen at the right and left renal arteries which is a little bit down from the aorta location so here's right okay none note it and then over the left then we're gonna listen at the iliac and it's a little bit below the belly button right here and this is Illya Carter II and then listen on the other side and again like I pointed out you could listen at the femoral artery and the groin if you need it too now we're going to do palpation first we're going to do light palpation then deep and being as I do this please tell me if you feel any pain or tenderness so first we're gonna do by palpation we'll just start in the right lower quadrant and work her way around and you're gonna go about two centimeters and you're just feeling for any rigidity any lumps masses anything like that how's that feel okay okay now we're gonna do deep palpation and we're gonna go about four to five centimeters so a lot more deep then again you're just feeling for any masses lumps and then tell me if you have any tenderness and sometimes you can do this with two hands if need be if you're not strong enough [ __ ] me telling anything feels nice and soft hurts um belly sounds that's why you do this after you listen because you stimulate it good so that wraps up how to perform an abdominal assessment and don't forget to check out that video on the complete head-to-toe nursing assessment thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	N95_Mask_How_to_Wear_N95_Respirator_Nursing_Skill_Tutorial.txt	hey everyone is cereth register nurse sorry and common in this video i'm gonna demonstrate how to put on and take off an n95 respirator before dawning or putting on an n95 respirator you always want to follow the manufacturer's instructions also you want to follow your facility's protocol for getting fitted for your n95 respirator and this is usually done yearly but it can vary first what you want to do is you want to perform hand hygiene hold the respirator in one hand with the front of the respirator touching the inside of your hand the metal nose piece should rest near the fingertips be sure the top and bottom straps of the respirator are hanging down below your hand and are not tangled or twisted take the respirator and place it over your nose mouth and under the chin hold the respirator in place with one hand use your free hand to place the straps on your head it's very important the straps rest on the head and that they are not overlapping or tangled first take the top strap of the respirator and place it just past the top of your head above the ears then take the bottom strap of the respirator and place it around the bottom of the head just below the ears again make sure the straps are not crossing or are overly twisted to ensure a proper seal of the respirator around the nose area use the fingertips of both hands to bend the metal nose piece around the nose starting in the center and working your way outward on each side now perform the seal test this is performed to check if you have donned the respirator correctly and that it does not leak air always follow the manufacturer's guidelines for this part because it can vary between brands first place both hands over the respirator then take a sharp breath in and make sure the respirator is sealing over the face during inhalation keep both hands on the respirator breathe out while breathing out make sure you feel for air leakage around the nose piece and around the respirators perimeter if leaking air around the nose piece try rebuilding the nose piece with both hands and recheck the seal if air is leaking around the respirators perimeter try readjusting the straps and recheck again make sure the respirator is not being affected by any piercings facial hair and glasses which prevent a proper seal if respirator still leaks after this you may need a different size or type of respirator before use now it's Adolphe or take off the mask two things you want to remember is that you don't want to touch the front of the mask because it's considered contaminated but you can touch the straps of the mask because these are considered clean first you will take your hands and you will remove the bottom strap by pulling it over the head then you will take the top strap and pull it over the head be very careful not to touch the mask now you can dispose or reuse the mask depending on your facility's protocol and then perform hand hygiene okay so that wraps up this video on how to dawn and dolphin n95 respirator and be sure to check out the other videos in this series
Nursing_Skills_Videos	Crutches_Canes_and_Walkers_Nursing_NCLEX_Assistive_Devices_Review.txt	hey everyone is cereth red sterner sorry and calm and in this video we're going to go over canes crutches and walkers and as always whenever you get done watching this video over assistive devices you can access the free quiz for exams in nursing school you want to make sure that you are familiar with assistive devices such as crutches canes and walkers and in our next videos we will be talking about canes and walkers so with crutches what you specifically want to know for exams is how they should properly fit the different types of gates you can do with crutches how a patient should go up and down the stairs with crutches and how they should get up or set down in a chair so first let's talk about proper fit what are some things that you need to see in your patient that tells you that those crutches fit them properly well before a patient even starts using crutches the crutches will have to be adjusted based on the patient's height and they can be adjusted at the bottom and at the talk and usually on most crutches there's a height range where you will slide the part of the crutch to meet whatever your patient's hype is also there's a hand grip on the crutch that can be adjusted as well so some things you want to remember is that you want to be looking at the gap between the crutch rest pad and axilla the armpit there should be a gap there it should be about two to three fingers width which ends up being about one to one and a half inches and the reason you want this gap is because while the patient is using using the crutch they need to be putting all their weight on the hand grips rather than on this crutch rest pad via that Scylla area because in this area you have nerves and if they're putting all their weight on that number one it's really going to be uncomfortable and hurt but it's also gonna damage those nerves in that area also you want to be looking at the hand grips and where they're lining up on the body the hand grips should be even with the top of the hip line so whenever the patient actually uses the handgrips their elbow will be slightly bent at about 30 degrees now let's talk about the different types of gates while using crutches and you want to make sure that you really truly understand these different types of gates that I'm going to go over because exams love to give you a description of a gate that a patient may be doing while using crutches and you'll have to identify it so before a patient even starts ambulating with crutches at first you want to make sure that they are wearing a gait belt for safety also before they start doing one of these type of gates they're going to start in the tripod position and it looks something like this and it forms like a triangle Hintz looks like a tripod and this is where the crutches are about 6 inches out diagonally from the feet so first let's start out with the point gates we have three of them we have the 2-point gate the 3-point gate and the four-point gate now how you can keep these straight is that you need to ask yourself how many points are on the ground hence how many crutches are on the ground and how many feet are on the ground whenever you're looking at the scenario because each crutch counts as a point and each foot counts as a point so the two point we're gonna have two points on the ground at a time whether it's a crutch or a foot so what does it look like well this is where the patient will move the crutch on the injured side so we're going to say it's the right side so they move the right crutch and they move the left foot together then they will move the left crutch on that non injured side and the right foot together so you have two points next is the 4-point gate and it's a little bit similar to the two-point gate but each point is moving separately because remember in two point they were moving together the Rye and the left moving together and then the left and the right moving together be here with four point they're separate so the patient will move the right crutch witches will say the injured side then they'll move the left foot then they'll move the left crutch and then they'll move the right foot and the last point gay is the 3.8 and this is where they move both crutches and the injured leg together at the same time and then they will move the non injured leg and then lastly we have the two swing gates and it's either a swing to gate or a swing through gate and let the names help you so the swing to gate is where the patient will move both crutches forward then they will hint swing or move both legs forward and place them at the placement of where the crutches are located now the swing through is very similar to this they will move both crutches forward then they will move both legs forward can't swing both legs forward but they will swing it pass the crutch placement now let's talk about stairs how does a patient navigate up and down the stairs while using crutches well you want to keep these two straight and what I'm meaning is which leg is going to go first up the step versus which leg is going to go down the step first and to remember that remember good up and bad down so whatever a patient is going up the stairs their good leg is going to go first up on the step followed by the crutches and the bad leg which will proceed and go up the step now whenever they're going down the steps they're going to move the crutches down onto the stab that will help provide stability followed by the Bab leg because a bad leg is going to go down and then they're going to move the good leg down on to the step and lastly we're going to wrap up the lecture and talk about how a patient sits down and gets up while using crutches to sit in the chair the patient's going to back up to the chair and fill the chair with the non injured leg and when the patient fills the chair with a non injured leg they're gonna stop and move both crutches over to the injured side for support then the patient's going to grip the hand grips and slightly bend the non injured leg and feel behind them and then set in the chair while keeping the injured leg extended to get up from the chair the patient is going to take the crutches and put them on the injured side for support he's going to keep the injured leg extended and push up on the non injured side and using the hand grips of the crutches then he's going to put the crutches in position okay you want to make sure that you're familiar with how a cane should properly fit a patient how to actually walk with a cane how a patient should go up and down stairs with a cane and how they should get up from a chair or sit down in a chair with the cane therefore let's start with the proper fit how do you know as the nurse of this cane actually fits your patient well before a patient uses a cane for the first time it has to be adjusted most canes can be adjusted at the bottom by sliding the cane into location of where it should go to fit the patient but once the patient is holding the cane or once they have the cane standing beside their body how do you know it actually fits the patient well there's two ways you can tell the first way is that the top of the cane which is felt this area here should be even with the great trochanter the great trochanter is a prominence of the femur so that's all but the cane should rest about right there whenever the patient's standing up and has the cane beside of them or the top of the cane should be even with the wrist crease that is closest to the hand so those are two ways you can tell also whenever the patient is clean the cane the elbows should be flexed at a 15 to 30 degree angle so those are some things that you can look for as the nurse to tell you that this cane properly fits your patient now let's talk about how a patient should walk with the cane before a patient starts using their cane and practicing with it you want to make sure that they are wearing a gait belt for safety in addition you're going to stand on the patient's weak side in case they lose their balance so whenever they start using the cane they want to make sure that they're in the proper position the position that they want to be in they want to make sure that they are positioning the tip of the cane at least four inches from the side of the foot and they want to hold the cane on the strong side of the body so remember that the cane needs to be on the strong side very important concept to remember so how are they going to actually ambulate with this cane well to emulate with a cane what they're going to do is they're going to move the cane with the weak side together forward so they move the cane along with the weak side together and then they will move the strong side forward so how does a patient go up and down the stairs with a cane well the concept is the same like how we learned with crutches remember I told you to remember up equals good down equals bad and what we're referring to is the good leg going up first which would be the strong leg versus whenever we're going down the stairs it would be the bad leg that's gonna go first so the weak side therefore how do we go up the stairs using that concept well what the patient wants to do is they want to hold the cane on that good / strong side then they're going to move the good leg up onto the step and they're gonna put weight onto the cane and then move the cane and the bad leg up onto the step now to go down the stairs the patient again is going to hold the cane on the good side that strong side they're gonna move the cane down onto the step with the bad leg so the bad leg is going down then they're going to move the good leg down onto the step and lastly let's talk about how a patient will sit down and get up using a cane to sit down with a cane the patient is going to back up to the chair until he fills the chair with the back of the legs then the patient will allow the cane to rest on the side of the chair and place both hands on the chairs armrest and place weight on the hands while keeping the weak leg extended out and been the strong leg to sit down to sit up with the cane a patient's going to place the cane on the strong side and lean forward in the chair while keeping the weak leg slightly extended forward then the patient is going to push down on the canes hand grip and the chair armrest and then put weight on the strong leg and stand in position with the cane when you're studying Walker's for your exams there are some things you want to remember you want to remember the proper fit what are those characteristics that the Walker actually fits your patient how to walk with the walker so the gait and how the patient to get up and sat down in a chair so first let's talk about proper fit how do you know that this Walker actually fits your patient well before a patient even uses a walker for the first time it has to be adjusted and on most Walker's they're adjusted down at the bottom you have to adjust each leg there's four of those legs so you've adjusted the Walker now what do you look for these are things you want to remember for your exams well have the patient hold onto the hand grips while they're standing with the Walker and you're going to look at that elbow there should be about a fifteen to thirty degree bend in the elbows also when the patient holds their arms down at their side their wrist crease should line up with those hand grips those two things really tell you that this Walker fits this patient the next thing you want to know about is the gait how does your patient actually walk with the Walker and this is something that you really want to pay attention to for exams so you want to make sure that you're paying attention to what's moving first is that the walkers at the weak side or the strong side and how that order goes so before a patient actually uses a walker for those first couple times you want to make sure that you have applied a gait belt to their waist for safety also you want to stand on the patient's weak side in case they start to stumble or fall you want to be there and you want to tell the patient before they start ambulating with their Walker that they don't want to stare down at their feet while they're doing it even though that just seems like something natural you want to do that can actually mess them up and cause them to fall they want to look straight ahead just like if they were walking normally and the starting position how do they start out walking with their Walker well they want to make sure that the back tips of the Walker match up with the middle of their foot now let's talk about how you actually ambulate with a walker to ambulate with the Walker first the patient's going to get in position and hold onto the handgrips of the Walker first they will lift and move the Walker forward and then make sure all four points of the Walker are touching the ground then they will move the weeks put weight on the hands the other hand grips and then move the strong side again they will lift the Walker move it forward make sure all four points are on the ground then they will move the weak side put weight on the handgrips and then move the strong side to sit down in a chair a patient is going to take their Walker hold on to the hand grips and slowly back up to the chair until they fill the chair with the back of their legs then they're going to slightly extend that weak leg out and take their hands and position them behind them and then their strong leg and feel for the chairs armrest and then set down to get up from the chair what the patient's going to do is of course make sure the Walker is out in front of them they're going to lean forward out of the chair make sure their hands are on the hand rest and slightly extend that weak leg out then they're going to put weight on their hands by pushing up on the armrest of the chair and with their strong leg and putting their hands on the hand grip of the Walker then they are ready to ambulate and again to do that they will lift the Walker make sure all points are on the floor move that weak leg put weight on the handgrips and then move the strong side thank you so much for watching don't forget to take the free quiz and to subscribe to our channel for more videos
Nursing_Skills_Videos	How_to_Prime_IV_Tubing_Line_How_to_Spike_a_IV_Bag_for_Nursing.txt	hey everyone it's s register nurse rn.com and in this video I'm going to demonstrate for you how to spike an IV bag and prime the tubing so what does Spike and prime mean spiking the bag means that we are going to use the spike from the tubing to penetrate the bag so we can withdraw the solution so it can go into the patient's bloodstream and priming the IV tubing means that we're going to take this solution and put it through our line to remove any air because we don't want to inject air into a patient's bloodstream because it can cause an air embolism so as a nurse you are going to be doing this all the time this is standard care so if you are a student or a new nurse um I really recommend that you get familiar with this skill and practice as much as you can so the whole goal of IV therapy is to take this bag of solution may just be basic normal Sal or a medication like heprin and we want to get this into the patient's bloodstream so it'll use tubing and it'll go through that tubing your pump will regulate it and then it'll go into the patient's bloodstream so that is a route of how we give medications okay first what you want to do is you want to perform hand hygiene because this skill requires that you use aseptic technique and we want to be very careful not to contaminate our bag or our IV tubing because we could cause the patient an infection and next what you want to do is you want to confirm that you have the right solution so you'll look at your physic's order look at your bag make sure it matches up and make sure that you have the right Patient next you'll want to get some IV tubing and every facility has different types of tubing depending on their supplier so get familiar with what you have but how you spike a bag and prime the tubing they all tend to work the same and also make sure that your bag comes with a label so you can label the tubing it's usually found sometimes staple to the bag inside the bag or you may have to pick it up in an Alco where Supply puts it but you will want this because it's very important that you label your tubing whenever you initiate new tubing because it expires and here are some key points you want to remember um with continuous tubing sets that you're going to use if you're not using it to um administer blood blood products or fat emotions like tpn or lipids the CD recommends that you change the frequency of the tubing every 96 hours which is 4 days however hospitals have their own protocols based on the cdc's protocols it's usually anywhere from 72 hours to 96 hours most of the places I've worked it's 72 hours so every 3 days now say you were going to be starting blood or a blood product or tpn or lipids um the CDC recommends that you change the tubing every 24 hours and that tends to be standard care for hospitals so always keep that in mind and the very first thing what I like to do is so I don't forget to label my tubing because it's really important because if another nurse comes in behind you taking over patient care they need to know when that tubing expires do they need to change it or they just going to have to throw the whole thing away because they're unsure and that waste supplies so I like to label my um little label first and so I don't forget because if I get really busy may get lost in the patient's bed fall on the floor and then I'll completely forget okay first we're going to fill out the start date and this is when you initiated it so go ahead and fill that part out so we're going to put March 10th 2017 and HR means hour and that's the time that you started it and this needs to be written in military time so we're going to put 0845 for 8:45 in the morning and then the discard date is when um this is according to your hospital protocol whenever they say that you need to to change it out so we'll say it's 72 hours so we'll need to discard this on 313 2017 and the hour will match up whenever you start it so 0845 and then put your initials now what we want to do is we want to open our tubing so just tear it open and take our tubing out and you want to unkink the tubing and remove any tape that's keeping the tubing together and what I like to do is I like to go ahead and apply my label to the tubing so I don't forget and the best place to do this is right below this drip chamber so whenever it's hanging up there on the pole everyone can see it and it's easy to find you don't have to look down or around it's just completely right there so we have it labeled now let's look at the parts of the IV tubing at the top underneath this cap is where you have the spot this part will go inside the bag to penetrate it so we can withdraw the fluids and you don't want to touch when this caps off that Spike because you will contaminate the tubing and you'll need to start all over below that this area right here is the drip chamber we will squeeze that when we start to allow the fluid to drop in there so we can get it to go through the line um as we go down we have these little access ports that is used to administer medications any IV pushes and as we go down through our tubing this part right here for this particular tubing set is for an IV pump and this acts as a key when it goes into the pump and you can set your rates um this part right here get really familiar with that because you will be using that a lot it's the roller clamp you just roll this part of the clamp to turn it off and we're going to keep it in the off position right now so whenever you always start turn it off because when we penetrate this bag we don't want all these fluids rushing out on us we want to be able to control it so it's in the off position right now and then this part right here is the part that will screw onto the patient IV central line whatever they have and we'll remove that blue cap and um whenever we remove the cap we want to be really careful not to touch the inside of that part of the part of that because um we can contaminate it now what we're going to do is we're going to spike the bag and whenever I do this I like to wear gloves because I don't want to get contaminated with whatever type of medications in here especially if you're going to be giving nitroglycerin or something like that you don't want to get that on your hands and I want to decrease the chances of me contaminating it myself so um what we're going to do is we're going to put this Spike inside the spike por now you have two ports on your IV bag you have a medication administration Port which we will not be using you use this port to instill medications into the fluids a lot of times Pharmacy does that and um you have right the side of it the spiking P so what we're going to do is we're going to hold the ivy bag in our non-dominant hand then we're going to pull this stopper off and then we're going to remove the cap from the tubing and we're going to be careful not to contaminate these by touching them and then we're just going to penetrate the spike into the spiking port and you really got to twist to get it in there you just it's not very easy you just can't slide it in there you got to really twist it okay and then we're going to turn it back and see nothing's flowing out because we have our clamper stopped so that's good and then we're going to hang it on the pole now we're going to squeeze our drip chamber and we want the fluid from the bag to enter into this chamber and to feel where this line is so we're going to squeeze it you got to give it a couple squeezes sometimes and that's about as much as we want now you're going to open up the clamp by turning it to this position and what's going to happen is that all that fluid from the drip chamber is going to drain through your tubing and this is what's called priming the line because we want to remove any air from the line and you will start to see it come out this area and this blue cap on here you'll probably have to loosen it a little bit so the fluid can come out so let's unclamp a roller clamp and I like to control the flow a little bit so it's not all coming out really fast here in a second it'll be coming and it'll start to drip out as you can see and you can let this drip into the sink or wherever you have access to let it drip and you want to let all the air bubbles get out you don't want any large air bubbles in your lime and our tubing is nice and primed now okay so you always want to just double check your line make sure there's no massive air bubbles in there so um here I just wanted to demonstrate for you so you can see what I'm talking about here's the fluid right here but from here to here is air we do not want that so um you can flick it a little bit and you need to ref flush your line just to make sure there's none in there um so it won't go to the patient so after you have primed your tubing you're ready to connect your bag to the IV pump which will control the rate of how fast your bag will drip in and you will connect the tubing to your patient this little blue cap twists off and you'll just screw this part into your patient's axess whatever they have and say that you're not ready to connect this yet um just make sure that you keep this part covered just so it doesn't become contaminated and put your roller clamp in the off position so your solution isn't spilling out everywhere okay so that wraps up how to spike an IV bag and Prime IV tubing thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	How_to_Go_Up_Down_Stairs_with_a_Cane_Nursing_Cane_NCLEX.txt	hey everyone this is hair thread sterner sorry and calm and in this video I'm going to demonstrate how to go up and down the stairs with a cane so how does a patient go up and down the stairs with a cane well the concept is the same like how we learned with crutches remember I told you to remember up equals good down equals bad and what we're referring to is the good leg going up first which would be the strong leg versus whenever we're going down this stairs it would be the bad leg that's going to go first so the weak side therefore how do we go up the stairs using that concept well what the patient wants to do is they want to hold the cane on that good slash strong side then they're going to move the good leg up onto the step and they're gonna put weight onto the cane and then move the cane and the bad leg up onto the step now to go down the stairs the patient again is going to hold the cane on the good side that strong side they're gonna move the cane down onto the step with the bad leg so the bad leg is going down then they're gonna move the good leg down onto the step thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	Intramuscular_Injection_in_Deltoid_Muscle_with_ZTrack_Technique.txt	this is our thread sterner sorry and calm and in this video I'm going to demonstrate how to give an intramuscular injection in the deltoid before ever administering any type of medication including ayam injections you always want to review your facility's protocol and how they require you to administer that certain medication now before actually giving the ayam injection there's some prep work you have to do first of all you have to make sure you have the right patient the right medication are you giving the right dose at the right time and the right route so double-check all those things in addition you want to get your supplies ready and to do this you have to do some critical thinking so first of all you need to ask yourself ok what muscle are we going to be giving this injection in here in this video we're going to be doing the deltoid so let's review some things about the deltoid muscle it's located up here in this region and this muscles a little bit smaller than the other muscles you can use it can only hold about one to two milliliters of a solution so if you need to give more amounts of a solution you need to go in a larger muscle like the ventral gluteal muscle next you want to look at how your patient is built generally the guidelines for a needle length in giving a deltoid injection is about a one-inch needle to a one and a half inch needle so look at your patient do they have a lot of adipose fatty tissue over that deltoid muscle if so you want to use the longer of the two options so one and a half needle length would be good here we're going to use a one inch needle next you want to look at the gauge of needle this is how big that hole of that needle is that allows that solution to go through so most vaccines are watery so you can give anywhere from a 20 to a 25 gauge needle here we're gonna use a 23 gauge but if you were doing like a thicker solution that's all you would want to use anywhere between an 18 to 25 gauge needle no what we're gonna do is we're going to gather supplies perform hand hygiene and it's optional if you want to wear gloves according to the CDC gov gloves are not required when administering vaccines unless the person administering the Sene is likely to come into contact with potentially infectious body fluids or has openly jhin's on hands therefore gloves are optional personally I like to wear gloves because I never know what I may come into contact with then of course you want to explain the procedure to the patient what you're going to be doing what you're administering to them in a tip whenever I give vaccines I always like to ask the patient are they right or left-handed because I like to give the deltoid I am injection and their non-dominant arm because if you've ever had a vaccine or an injection before in your arm it can get sore so always try to take that into consideration before you give the injection you can have the patient stand or sit if your patient doesn't like needles it's probably best to have them sit in case they pass out so where we're going is the deltoid muscle it's located up in this area so what you want to do is you want to have the patient completely like relax their arm you do not want this muscle to be tensed up because it's going to be a little bit more painful for them so tell them to relax as much as possible then we're going to find where we're going so to find the deltoid muscle and for an injection site we want to use landmarks so our first landmark is called the pro Mian process this is found high up where the shoulder is and it comes out it's like a bony prominence just fill in yourself where it is it's really easy to find so you're gonna feel that and you're going to go about two fingers width down from that area so we have our two fingers so we're going to go about right here in this area once we have our area what we want to do is we want to clean the injection site so we're going to take our alcohol prep we're going to start in the center and we're gonna work our way outward and cleanse the area and let that dry completely don't blow on it let it air dry now to give the injection what we're gonna do is we're gonna use what's called a z track technique this is now recommended for all I am injections and what this technique does is number one it decreases pain for the patient but it helps prevent the solution that we're instilling from actually going in that sub cute issue we want this to go in the muscle not the sub-q tissue so the z track technique helps with that back in the day they taught to pinch the skin up bunch it up to give the vaccine that's no longer recommended but the Z track method so what we're going to do to do that we're going to take our non-dominant hand and we're just going to go to the side of where an injection site is going to be and we're just going to put a little bit of pressure and we're just going to pull the skin to the side then we're going to take our dominant hand with injection the needle and we're going to go in like we're shooting a dart at a 90 degree angle and you'll want to do this quickly to cause the patient less pain so we want to steady our syringe so we're going to take our thumb and our forefinger and just hold it steady then we're gonna take our dominant hand and depress the plunger and instill that fluid slowly about over 10 seconds per ml this was half an ml so we're gonna do about five seconds and then once you have instilled it wait about ten more seconds to let all that fluid go down into that muscle and then what we're gonna do is we are going to take it out the same angle we had inserted it we're going to engage our safety never recap a used needle then you can take a gauze and cover the area sometimes it can bleed if your patients on blood thinners but never massage the area because that can increase the solution going into the sub-q tissue and then dispose of your syringe a needle in the sharps container once you're done with that of course what you want to do is you want to perform hand hygiene and you want to document you're going to document what muscle you use on what side and how the patient responded along with how much medication you instilled in that muscle now notice whenever I was giving that I am injection I did not aspirate aspiration for ayam injections is no longer recommended a core to CDC gov they say aspiration before injection of vaccines or toxoids which is pulling back on the syringe plunger after needle insertion before injection is not necessary because no large vessels are present at the recommended injection sites and a process that includes aspiration might be more painful for infants okay so that wraps up how to give an iamb injection in the deltoid thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	How_to_Read_a_Syringe_3_ml_1_ml_Insulin_5_mlcc_Reading_a_Syringe_Plunger.txt	hey everyone the Sears register nurse re and calm and in this video I'm going to go over how to read syringes in this video I want to go over five basic syringes that you will be using as a nurse and show you how to read each of them now as a nurse you're going to have a lot of different types of syringes at your disposal and you want to select the correct syringe for the amount of medication that you want to get for instance say that the physician has ordered 0.5 milliliters of morphine the 3 cc syringe is the best for drawing that medication up to get the most accurate dose however on the other hand if you're going to give point 2 5 milliliters of a medication the 1 milliliter syringe is the best for that so let me show you the basic parts of a syringe here is a basic 10 milliliter syringe and anytime you have a syringe you always want to look at its total capacity it can hold and this one holds 10 MLS now as a side note 10 MLS is the same as 10 cc's so if you hear someone say let's give 2 cc's it's the same as two MLS just to let you know that now let's go over the basic parts of the syringe okay so here at the top this is our adapter part and this is where you would screw on the needle if you were going to give an iamb injection and now in the hospital we use needle list devices so we would just screw this adapter part on to the IV the hub of the IV or the line and then we would just give our medication that way now the other part that's important is the barrel of the syringe the barrel has a scale on it and this scale tells us how much of a medication we're gonna give based on whatever the doctor ordered now every syringe is different and how it measures that's why I say get familiar with how much your syringe holds which we're going to go over in depth here in a second with each individual syringe and you have a scale so at the top it's zero then you have a line in between that and then you have one then you and then you have a big - and it measures onward now one thing that you want to keep in mind is how to actually measure the fluid that you're drawing up with this plunger so let's go over the other part the plunger I'm going to take it out so you can see it the plunger has a top part and a bottom part and it also has a beveled area and whenever you're actually drawing up the medication you're going to line the line up that's on the scale of your barrel with this top part of the plunger not the beveled part the top part now let's go over how to read each syringe first we're going to start with the easiest syringe which is the 10 milliliter or 10 cc syringe and as you can tell on this picture each area is measured out so the top part where the top line is 0.5 milliliters then it goes to one then the next line is 1.5 then it goes to two then the next line is 2.5 and so on so this syringe based on how its scale is set up in its capacity it measures by 0.5 now let's test your knowledge based on the yellow wine what is the measurement of this syringe and the answer is 4.5 MLS now let's look at the five milliliter syringe and as you can see on this this has a little bit more lines in between the main measurements than the 10 cc syringe and that very top line is zero and then below zero you have 0.2 then point four seven point six point eight and then the huge line is one so that is where one milliliter is at then after that you'll count by twos the point twos so to be 1.2 1.4 1.6 1.8 and 2 and then so on it would keep measuring that a male until you hit 5 MLS now let's look at this syringe based on the yellow wine what is the measurement of this syringe the answer is 2.2 MLS now let's look at our 3 no leader syringe this syringe even further breaks down the measurement so you can really draw up a smaller amount of a medication for instance like I said at the beginning of the video this syringe is really good for drawing up a medication if you just have to give point five of something so that top line is zero then as you pull the plunger down this syringe goes down by point one so you have 0.1 ml 0.2 0.3 0.4 and then you have half 0.5 then you have point six point seven point eight point nine and then one and then so on until you get a total capacity of three MLS based on the yellow wine what is the measurement of this syringe the answer is 0.7 MLS our next syringe is the one milliliter syringe and this syringe if you don't have good eyesight you really have to squint to see the little lines but it's really good for giving those medications where you have to give just a little bit like the 0.25 MLS okay on this syringe as you can see and the top line will be zero and then this particular syringe will measure down by 0.01 so you have point zero one point zero two point zero three onward until you hit point one ml's and then it'll just keep on going until you've hit a total capacity of 1 in MLS based on the yellow wine what is the measurement of this syringe the answer is 0.25 MLS okay this is our very last syringe and this is a syringe that you will be using a lot whenever you are giving diabetics insulin and it is a syringe that holds a total of a hundred units because insulin is measured in units and it looks a little similar to our one milliliter syringe that we went over but it's not measured in milliliters it's measured in units so as you can see with this we have our top line which is zero and this is going to measure in increments of two so the next line would be two units then four units six units eight units and then ten units and then onward until you would give a total of a hundred units based on the yellow wine what is the measurement of this syringe the answer is 72 units okay so that wraps up this video on how to read the common syringes you will encounter as a nurse thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	How_to_Take_a_Temperature_Under_Arm_Oral_Ear_Rectum_Skin_Temporal.txt	hey everyone it's sarah thread sterner sorry and calm and in this video I'm going to show you six different ways on how to check a temperature first let's talk about what is a normal body temperature in the adult it can vary anywhere between 97 degrees to 99 degrees Fahrenheit with the average about ninety eight point six and a child it can also vary anywhere between ninety seven point four to around a hundred degrees Fahrenheit and anything greater than a hundred point four is considered a fever in both adult and child the first way we're going to check a temperature is through the mouth so orally and you can do this one of two ways you can use a digital thermometer which is most commonly used or you can use a glass thermometer which really isn't used as much because they can break so let me show you first how to use this glass thermometer before taking an oral temperature you want to make sure that the person hasn't consumed anything and if they have wait about fifteen minutes before taking the temperature so what you want to do is you want to perform hand hygiene and dawn gloves anytime you're taking an oral temperature you may be encountering oral secretions say you want to prevent yourself from getting contaminated then you want to make sure your thermometer is clean so clean it with alcohol prep and let it dry now with glass thermometers you have to pay attention to where the liquid is on the reading scale if it is above 98.6 degrees Fahrenheit you will have to take your wrist and flick the thermometer down for that fluid to go below ninety eight point six degrees once the fluid in the thermometer is below that point you'll want to insert the thermometer in a probe cover just to protect the thermometer from being contaminated any more and then take the thermometer and insert it in the patient's mouth have the patient lift up their tongue and put it underneath the tongue okay keep it closed and wait for three minutes and then remove the thermometer once the time's up you'll take the thermometer out of the patient's mouth and read the thermometer then what you'll want to do is remove the probe cover clean your thermometer with alcohol prep or whatever your facility wants you to clean it with Dolph your gloves perform hand hygiene and document the temperature especially the route now let me show you how to check an oil temperature using a digital thermometer of course what you want to do first is perform hand hygiene and dawn and gloves and make sure you're working with a clean thermometer so always clean it with alcohol prep then you're gonna take the thermometer and insert it in its probe cover then turn the thermometer on by hitting the button on the thermometer and insert it in the patient's mouth have the patient lift up the tongue and then put it over and then keep the mouth closed and then once the thermometer is done it will beep once the thermometer beeps take the thermometer out of the patient's mouth and read it then when you're done taking the temperature be sure to remove the probe cover clean the thermometer / your facilities protocol with alcohol prep and doff your gloves perform hand hygiene and document the temperature another way you can take a temperature is through the ear via the tympanic membrane so first what you want to do of course is perform hand hygiene get your thermometer and with these type of thermometers you want to use a probe filter so place that just over the thermometer turn your thermometer on and you're going to insert this in the ear for the adult what you'll want to do is you want to take the pin out of the ear pull it up and back if your patient was less than 12 months of age you would want to take the pin up and pull it down and back so we're going to pull it up and back we're going to insert the thermometer in the ear then we're going to hit the button to take the temperature just takes a second then we will read our thermometer and it will display the temperature right here then after reading the temperature what you want to do is take the probe filter off and perform hand hygiene and clean your thermometer / your facilities protocol and document the temperature a third way to take a temperature is through the armpit back Silla so to do that what you want to do is get a digital thermometer perform hand hygiene make sure your thermometer is clean and insert it into a probe cover then turn the thermometer on and lift up the person's arm place the tip of the thermometer directly into the armpit not on the clothing have them close the armpit and hold the thermometer there until it beeps once the thermometer phoebs take it out and read it after reading the thermometer reading take the probe cover off clean the thermometer perform hand hygiene and document another way to take a temperature is through the temporal artery so to do that of course we want to perform hand hygiene get our thermometer and with this specific thermometer it has a cap cover so take it off it does not have the little filters that go over it some models do the manufacturer just recommends cleaning the thermometer so make sure your thermometer is clean with alcohol prep or whatever your facility requires you to do what we're going to do is we're going to hold in this button keep it held we're going to go to the middle of the forehead and swipe a thermometer across the forehead to the hairline and then release the button and read our temperature now let's say that the patient was sweating had a lot of sweat on their forehead how we'd want to do that is again hold the button in go through the middle of the forehead sweep across to the hairline and then go behind the ear and read our thermometer then clean your thermometer recap it with the protective cover perform hand hygiene and document another way to check a temperature is rectally so to do that you want to perform hand hygiene and dawn gloves when checking the temperature this way you want to get a thermometer and you want to make sure it's clean then insert the thermometer into a probe cover and then lubricate the tip of the thermometer then turn the thermometer on and have your patient positioned on their back with their knees bent then insert the flowmeter into the rectum about one inch then remove the thermometer once it beeps and read the temperature measurement after reading the temperature take the probe cover off clean the thermometer off your gloves perform hand hygiene and document another way to check your temperature is through the skin using a temperature strip and to do that you want to perform hand hygiene peel off your strip and put it on the forehead and after placing the strip within 15 seconds a dot will start to appear and whatever turns green will be the temperature reading then once you have read the temperature just take the strip off and dispose of it then perform hand hygiene and document now you may be asking yourself which route is the most accurate according to the experts the rectum is the most accurate route to take a temperature the least accurate route would be the skin or the armpit so always use that as last resort and whenever you take a temperature in the rectum it's always going to read about a degree higher than the mouth and if you take a temperature in the armpit it's gonna read about a degree less than the mouth thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	Lung_Auscultation_Landmarks_Sounds_Placement_Nursing_Assessing_Lungs_Part_1.txt	hey everyone it's Sarah register nurse rn.com and in this video I'm going to be going over lung oscilation specifically the sites of where you osculate and I'm going to be going over normal breath sounds versus abnormal breath sounds and as always you can access the quiz and the notes over here or in the description below now in the next video I'm going to be performing an assessment on a patient and show you how to listen with your stethoscope to these sides so let's get started first let's talk about our objectives of this lecture what I want you to learn is lung Anatomy because it's really important you know what you're listening to the specific oscilation sites I'm going to give you some landmarks to make your job easier so you'll know what intercostal space correlates with which lobe of the lung um we're going to go over some audio clips of normal breath sounds and where you should hear them at CU that's a big thing of you should know where you should hear bronchial versus bicular and we're going to go over abnormal breath sounds and I'm going to let you listen to them and you can compare the two so first let's go over lung Anatomy whenever you're osculating you're going to be listening to the anterior part of the chest and the posterior part of the chest now one thing you want to remember whenever you're listening to anterior mostly what you're going to be hearing a lot of because it's predominantly in the anterior part are the upper low loes however on the posterior part what is predominately made up on this is the lower loes so first let's take a look of the anatomy of the anterior part of the chest in this illustration you have your right lung and your left lung and how I set up this illustration is that I wanted you to be able to see what is over the lungs because whenever you're listening with your stethoscope you need to know where your clavicle is and certain intercoastal spaces because they correlate to which lobe of the lung you're listening to for instance if you're listening down in the six intercostal space mid axillary you're assessing the lower loes of the right and left lung which we'll go over in depth whenever I cover these osculation sides so first let's cover the right lung your right lung is made up of three loes very important so you have your right upper lobe then you have your right middle lobe and right here here you have the horizontal Fisher and then you have the oblique Fisher and this just separates the middle loow from your upper and your lower and then down here you have your right lower then over here you have your left lung you have your left upper lobe and then your left lower lobe and if you notice in the drawing mainly whenever you're listening to your lungs you're mainly going to be assessing because you can see more of it of your upper load then we have our trachea which is up here and it goes down and it branches off into your broncus your bronchi and then your broni go even further and Branch off into bronchials with your avolar Sachs and this is where gas exchange occurs and then of course you have your breast bone and all your ribs right here now let's look at the posterior view of the lung here is the posterior view of a patient's back if you were to look inside look at the lungs you'll have the clavicle on the front then it comes around and it forms into your scapula and you have your left lung and your right lung they're on they're flip then compared to the anterior View and here as you can see you have mostly lower lobe that you're assessing and it's separated by your fissures here so you have your right upper lobe and you have your right lower lobe notice you can't really assess the middle lobe like as you can with anterior View and then over over here you have your left upper lobe and your left lower lobe and you have your spine in the middle and um it's really important you find C7 to T10 whenever you're assessing which we'll go over in the oscilation sites because this will help you know where to place your stethoscope in between the spine and the scapula on those intercostal spaces right in between the ribs before you start listening to a patient's lungs let's talk about some tips to make your job easier so you can get the best sound possible and be able to assess those lungs correctly okay first tip listen directly on the patient's chest with the diaphragm of your stethoscope that is the big part of the stethoscope which you'll see a little bit later and the reason you want to listen directly on the skin and not over cloes is because a lot of times your diaphragm can rub up against the patient's clothes and that can make like a rustling sound sound like one of those adventitious lung sounds and it'll muffle and decrease the sound quality of what you're trying to hear also whenever you are listening on women you want to have the woman raise up her breast so you can get underneath those sights so you can hear those lung sounds because the tissue will muffle the noise and you won't be able to hear that so remember that whenever you're listening on a female patient also whenever you're listening to the chest you're going to listen to the both the front and the back side and whenever you're listening you're going to note a full cycle of in inspiration and expiration and what you're listening for is you're listening to the pitch is it high is it a medium or low pitch what's the sound quality and its duration is is inspiration longer than expiration or vice versa or are they equal which are characteristics of normal breath sounds and and um on that sound that you're hearing is there any extra adventitious sounds that shouldn't be there maybe on inspiration or expiration so you really want to not that and whenever you listen you are going to start from the top and work your way down and you're going to compare sides so be something like this you're going to start up here at the apex of the lung listen here and then you're going to go over and listen on the other side to compare it then you're just going to drop down a little bit listen here and then you're going to go over here and compare sides drop down again and just keep repeating until you've reached the bottom another thing what you want to do to help get the best sound quality is have the patient sitting up so you can get to the front and the back of that chest posteriorly because this is an area that may give you trouble because as we went through the anatomy you have your spine here and you have the scapula here so if you have the patient sort of move their arms forward maybe in their lap to separate those shoulder blades you can get in those little intercal spaces better so you can put your diaphragm and listen to those sounds also whenever you're having the patient breathe you want them to breathe in and out through their mouth slowly so you can hear those lungs inflate and deflate however um a lot of patients who may have breathing difficulties you'll have to take your time with them because they can hyperventilate easily and and make sure your patient doesn't get dizzy and just take your time while you're having them breathe now let's go over the oscilation side okay first we're going to assess the anterior part of the chest first and what I like to do is I find the clavicle and um we're going to start at the apex of the lungs the top of the lungs and we're going to get our diaphragm which is the big part of your stethoscope and you are going to place it right slightly above that clavicle where the Apex is you're going to listen there for a full inspiration and expiration and then you're going to go over and compare on the other side and this is listening to the apex of the lungs then you're going to find your second intercal space this is one of those landmarks we were talking about because this is going to assess our upper loes of our right and left lung and this is found mid cularly so the middle of where the clavicle is in the second intercostal space so you will listen here compare your side and then just go a little bit lower maybe into the third intercal space and just keep listening to those upper loes then we're going to go down to our fourth intercal space this is another big Landmark for specifically the right middle lobe so we're going to go down to our fourth intercal space we're still in the left lobe in the upper lobe and now we're going to go over here and compare sides now we're in the right middle lobe and this again is mid perly and we're listening in here and then we'll just inch a little bit down maybe in the fifth and still assess our right middle lobe and then we'll go over and compare sides still being in the upper lobe on the left side now we will inch down to the sixth intercostal space but mid axillary so where their armpit is go Midway and we're down in the lower lobe of the lungs and we'll have them in inhale and exhale and then we'll go over and compare it on the other side which we're in the right lower lobe here and assessing this and then we'll just inch a little bit down maybe to the seventh space down the lung and just listen in those lower loes a little bit more and compare sides and then we're done now let's look at our sides on posterior just like with anterior in the posterior we're going to start from top to bottom and compare sides and work our way down and we're going to start right above the scapula right where the Apex is and we're going to listen here and then we're going to go over to the other side and compare and remember to get the best sound quality so you can hear so you're not listening over the shoulder blades because that will muffle your sound and you won't be able to hear have your patient put their arms in their lap or just separate those shoulder blades from each other so you can get in between that spine and shoulder blade area then what we're going to do we want to assess first your upper loes so from C7 to T3 your cervical and thoracic spine that is where your upper loes are and as you can see here's your Fishers right here and you have the upper loes there so what you want to do is just go in between where the shoulder blades and the spine are and just work your way down so we're going to listen to our upper Li so we'll go here and then we'll compare over here and then we'll go down a little bit towards where T3 is listen here still being in the upper lob then we'll go over and compare sides now from T3 to T10 that will allow us to assess our lower loes so we'll start around T3 and work our way down again just staying in between where the scapula and the spine is and we will just compare sides and inch our way down and you want to move around almost mid aill where you were moving before on anterior so you can just get a good feel for what's happening in those lower loes first let's start out talking about normal breath sounds okay there's three different types a tip for whenever you're trying to learn these normal breath sounds is to get a stethoscope listen to yourself or listen to others and get a rhythm down for how long inspiration expiration is and where these are located because that's the key with these three different sounds because they're heard in different areas throughout the lung field so let's go over them the first one is bronchial this is heard anteriorly only you're not going to hear this posteriorly anteriorly why because they are mainly hurt over the tracheal area with the stethoscope so up here in this area they are high pitched and loud and you will notice when you listen to them that the inspiration will be slightly shorter than the expiration and this is what bronchial breath sounds sound like next is bronchovesicular these are heard both anteriorly and posteriorly and they posteriorly you will hear these at the first and second intercal space so about in this area right in here with your stethoscope is where you're going to hear them anteriorly now posteriorly you're going to hear them in between the scapula so about right here where T3 T4 and the small little areas where you will hear those and they will have a medium pitch to them and inspiration and expiration will be equal and here's what Broncho visic sounds like and the third breath sound is called vesicular this is heard again an both anteriorly and posteriorly and it is heard throughout the peripheral lung Fields so you're going to be hearing these all throughout in this area over here anteriorly and posteriorly and it will have a low pitch that will be sort of soft and inspiration will be greater than expiration and here is what vesicular sounds like now let's talk about those abnormal breath sounds that you could hear that may be thrown in with those normal breath sounds okay they are separated into continuous and discontinuous now first let's go over continuous what does continuous mean this is a extra sound that you're hearing that is lasting SE more than 2 seconds with a full respiration okay the first type is called a high pitch polyphonic whe let the name help you okay so what is it it is mainly heard in expiration so when the patient's breathing out but it can be in Inspiration as well and it is a high pitch musical instrument sound with many different sounds to it that's why it's polyphonic and this is what a high-pitch polyphonic whe sounds [Music] like another type of Wee you can have is called a low pitch monophonic wheeze and this is again heard mainly in expiration but you can hear it any time A lot of times you'll hear it whenever the patient's breathing out it is a low pitch whistle so instead of being high pitch like the high pitch wh it's going to be low and it's going to be made up of one sound quality that's all you're going to be hearing and it can sound like a whistle or a whine and this is what a low pitch monophonic wi sounds [Music] like and the third type of continuous adventitious breath sound is called stri spider and this is heard on inspiration because what's happening is that the airway is being obstructed by inflammation or some foreign object something like that and once you hear this you will never forget it's very unique sounding it is a high pitch whistling or gasp with a very harsh quality to it and um patients like pediatric patients if they get the cro or cute epiglottis or you have a patient who has an airway struction you will hear this sound and this is what Strider sounds like now let's go over the second type of breath sounds abnormal breast sounds called discontinuous This is an extra sound that you're hearing that is La lasting less than2 seconds okay first type um is coarse crackles crackles for nor Al has been known as rails so if you hear that that's what it means crackles rails they're interchanged just like ronai and Weis so course crackles they are mainly heard in Inspiration when the patient's breathing in and can extend into expiration and what it will sound like is a low pitch wet SL bubbling sound and this is what coar crackle sounds like the second type of discontinuous abnormal sound is called Fine crackles this is heard on inspiration and it is a high pitch crackling sound compared to the coarse crackle this is like low pitch like a bubbling noise this fine crackles is high pitch it sounds completely different than coarse crackles and it has like a crackling a fire sound to it and the key with this is that it does not clear when you have the patient cough so you listen you hear that you ask patient to cough and it's still there that would be fine crackles and this is what it sounds like and the last sound is called a plur friction rub this is heard both on inspiration and expiration and it is a low pitch harsh grading sound and what's causing this is that your plora on your lungs those two layers are rubbing against them each other and they normally have this little thin layer of Cirus fluid around the lung but it doesn't right now due to all that inflammation going on so you can actually hear that when that patient's breathing in and breathing out so that's why you're hearing it on inspiration and expiration now it can sound similar to a parac cardial friction rub how do you tell the difference um if you are wanting to know is this the lungs or is this the heart just listen have the patient hold their breath and if you can still hear that harsh grading sound it's the heart because they're holding their breath their lungs aren't moving so you've rolled out the lungs and this is what it sounds like okay that is lung occultation and normal breast sounds versus abnormal breast sounds don't forget to take the free quiz on the website and check out the other videos in this series to help you with lung oscilation and thank you so much for watching and please consider subscribing to this YouTube channel
Nursing_Skills_Videos	Suture_Removal_Nursing_Skill_How_to_Remove_Surgical_Sutures_Stitches.txt	this is cereth westerner sorry and calm and in this video I'm going to demonstrate how to remove surgical sutures specifically simple interrupted sutures sutures which are also called stitches are used to close a wound that could have been sustained through like an injury or a surgical procedure now there's various types of surgical patterns and in this video I'm going to demonstrate how to remove the most common type of surgical pattern which is the simple interrupted suture and this is where their individual sutures used to close the wound in addition as a nurse you can see various types and here you will see pictured a simple continuous suture which is just as the name says a continuous strand of suture closing the wound in addition to a vertical mattress suture a horizontal mattress suture a continuous blanket suture also called a forward interlocking suture and finally an interrupted cruciate suture as well and cruciate just means cross-shaped now keep in mind not all sutures need to be removed some suture material is made out of a dissolvable substance where I'll actually dissolve in the body itself while others have to be manually removed which is what we're going to be doing in this video in addition depending on where the suture is located and the depth the wound that the suture is closing the suture can stay in for as little as five days versus two weeks first what you want to do is you want to verify the physician's order because as a nurse before you can remove sutures you have to have an order next what you want to do is explain the procedure to the patient and get their verbal consent a lot of patients ask is this going to be painful no usually it's not a lot of patients report a tugging or pulling sensation but you can always give the patient pain medication if ordered prior to the removal of the sutures then gather your supplies one thing you'll need is a dressing change tray and before you actually remove your sutures you'll always want to follow your Hospital protocols because some hospitals say use sterile gloves to remove the sutures while others say you can just use clean gloves in this video we're going to demonstrate using sterile gloves so our sterile gloves will come here along with our sterile drape to put our supplies on and some gauze and our antiseptic to actually clean the suture line before and after the removal then of course you're going to need a suture removal kit and this comes with our tweezers our scissors to remove to cut the suture and some gauze to help us keep them in place and then you'll need some steri-strips and the steri-strips will be placed on the site where the individual suture was removed because one complication of suture removal is wound dehiscence and this is where your womb will open up prematurely and putting these on will help prevent that then you're going to perform hand hygiene and dawn clean gloves because first of all we want to do is we want to assess our womb before we remove the sutures so you're going to remove any old dressings that are present and first what you want to do is you want to check for infection is the site really red is it warm to the touch are there any hard areas is it using foul looking drainage or has a smell next you want to look at the suture line itself does the skin look nice and fused together or does it look weak like if you remove one of these sutures that wound is going to pop open and if any of that is present you want to notify the physician before actually removing the sutures then we're going to doll for gloves perform hand hygiene and a prepper supplies first what we're going to do is we're going to open up our dressing change tray and whenever you open these up on the top is usually the sterile drape which we're going to easily just grab and lay over here and the sterile gloves which we're going to lay right here and what we're going to be doing is opening up this sterile drape and we're going to drop our suture removal kit supplies and our steri-strips onto the sterile drape because we're trying to keep everything as sterile as possible to help prevent infection so what we're going to do now is open up our drapes so we're just going to gently open it and we have about two inches on the inside to be able to reach so we can open it up now let's open our supplies and drop it on to our field so first what we're gonna do is open our suture removal drop that in there and then discard that and then take our steri-strips and do the same now we're going to dawn our sterile gloves and if you don't know how to put on sterile gloves please watch my video on that so what we're gonna do is pull these tabs open lay these down and I'm gonna glove my right hand first cuz I'm right-handed so I'm going to grab this cuff and pull it over my hand and pull the cuff down then I'm going to take this hand which is sterile and slide it underneath the cuff part of this and I want to be careful not to touch my other hand okay now what we're gonna do is open up or antiseptic because we're going to use this to clean the wound before suture removal and after this helps decrease infection so let's open these up and inside you will have three swabs that have antiseptic on them so now take the antiseptic swab and clean along the wound this is going to prevent infection and then what we're gonna do is discard the swab and let the area dry now that the area is dried we are ready to remove the sutures so we're going to take our galls and we're just going to set it beside of our work area because this is where we're going to be dropping each suture and we want to dispose of those properly according to your hospital protocol because it's a biohazard and you want to count how many sutures you're removed and analyze them make sure the whole suture thread is intact and document so we're going to take our tweezers and put those in our non-dominant hand and we're going to take our scissors and put those in our dominant hand and what we're going to do first is we're going to remove every other suture starting at the second suture why is that because we want to prevent this wound from opening up prematurely so doing that will help hopefully prevent that and then after we remove every other suture before we remove the other sutures we're going to put steri-strips just to help reinforce that wound even more so two things you want to remember prior to removing sutures this is a really big thing you got to keep in mind where you actually cut the suture thread we're not going to cut on this side of the knot why is that if we cut that side of the knot when we go to pull that knot out this part of the thread that's been in contact with the outside environment has germs on it it's going to slide underneath that wound introducing all those germs into that wound leading to possible infection so we want to cut on this side of the knot that's closest to the skin so always remember that with these simple interrupted sutures in addition when we're actually removing the thread when we grab the knot we don't want to grab the knot and pull the suture thread away from the wound why is that if we pull away that's going to create tension on this incision line possibly leading to opening up so when we actually pull the thread up we're going to lift up gently and then we're going to pull over the wound starting at our second suture what we're going to do is we're going to grasp the not with our tweezers then we're going to take this scissors and we're going to just cut that part of the thread then we're going to lift up and pull over the wound and then just look at your suture make sure it's intact it hasn't fell apart or anything like that then drop it into your gauze then we're gonna do the same thing with the other suture so grab the knot snip underneath it then lift up and then pull over the wound and then just check your suture thread and drop in the ghost then take a new antiseptic swab and clean the areas where you remove the suture and let that dry then we're going to apply our steri-strips in the area where we removed each suture and you'll want to cut your steri-strips where they're at least 3/4 of an angel' each side of the incision and space them about 1/8 inch apart so we're going to just take our steri-strip and line it up where we have about 3/4 of an inch on each side and whenever you're applying steri-strips you want to gently lay down one part of it smooth it down and then gently lay down the other side don't pull it or tug on it to create tension because series strips are strong and they can actually tear the skin then we're gonna do the same thing with our other part line it up smooth it down and then gently lay on the other side now we're going to remove the remaining suture so again just grab your knot cut underneath pull up and then pull over and analyze your thread looks good and then placing the gauze again repeat the same process for the next suture grasp the knot cut underneath and lift up and pull over look at it looks good and drop it in the gauze and then our last one same thing grab the knot underneath lift up a little bit and then pull over look at your suture looks good now let's say while we were removing it this was starting to open up prematurely we would stop what we were doing don't remove any more sutures cover it up with a sterile gauze and notified the physician immediately then we're going to take a new antiseptic swab and we're just going to clean those areas again where we remove the sutures and we're gonna let that dry the side is dry so now we're going to apply our steri-strips the same way we did before leaving 3/4 inch on each side of the incision about 1/8 inch apart higher next one and then apply our last one same way after doing that you'll want to look at the site and ask yourself is this an area that's going to easily have friction on it like rubbing up against jeans or experience a lot of moisture if so you'll want to apply a dressing give the patient dressing supplies to change it regularly educate them about signs and symptoms of infection and winds who expect these steri-strips to fall off which is usually within ten days and tell them to let them follow all by themselves and do not remove them then after that what you want to do is you want to Dolf your gloves perform hand hygiene and document okay so that is how you remove surgical sutures thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	Denture_Care_Nursing_CNA_Skill_How_to_Clean_Dentures.txt	hey everyone it's sarah thread sterner sorry and.com and in this video I want to go over how to clean dentures as a nurse so as a nurse you're gonna be providing care to patients who wear dentures and your role will include helping the patient take those dentures out cleaning the dentures and providing mouth care because a lot of times patients are really sick and they can't get out of bed and do this like they normally would at home so you'll help them at night with doing that because they don't sleep with the dentures at night and you'll help them in between the meals after they eat and some things you want to keep in mind whenever you are taking care of dentures is that you don't damage them because they're very expensive and they don't become lost because you have a lot of people going in and out of the patient's room the patient's sick they're not really keeping track of where their dentures are so you'll as a nurse you'll want to keep them secured so first what you want to do is you want to gather your supplies which will include a denture cup which you'll want to put a patient label on the denture cup on the lid or write the patient's name the room number so they don't get confused with any other person's dentures also a soft bristled toothbrush with cleaning pace for the dentures it's really recommend that you use a denture cleaning pace instead of toothpaste because that can be abrasive to the dentures a kidney Basin towels paper towels and gloves and also be sure that you ask the patient if they brought any of their denture cleaning supplies because some patients may prefer that you use their supplies then you'll want to perform hand hygiene and dawn gloves and then assist the patient with removing their dentures so patients remove their dentures and we put them in the denture cup with the lid on it and we dr. gloves and perform hand hygiene and now we're at the sink area so first what I like to do is I like to prep my sink area because I don't want any of my supplies to become contaminated so since we're going to be turning the faucet on and off the lot we're going to put some extra paper towels there so we can just not contaminate our gloves and we can use the towel to turn it on and off then we're going to lay down a towel and this to help protect our supplies so let me be doing that just lay it down nice and neat and what we're going to put on it is our toothbrush with our bristles up our cleaning pace some extra paper towels to dry the Basin out with our basin of course our dentures in our denture cup and our gloves then we're going to prep the inside of the sink this is really one of the most important steps you want to take when you are cleaning dentures and what we're going to do is we're going to create this like cushion in the bottom of the sink because just in case if you draw up the dentures and let me tell you from cleaning a lots of dentures they become slippery and they can easily fall out of your hand so what we're going to do is we're going to just lay these towels down in the sink and then we're going to turn on our water but first let's pull the stopper and then just fill it up half way with water and again this just provides a nice cushion just in case those dentures were to fall into the sink okay and it's about halfway so we're gonna go ahead and turn the water off now we're going to prep our toothbrush and again you want to use like a soft bristled toothbrush or check with your patient or your supplies and see if you have a denture cleaning brush which is specifically made for cleaning dentures and we're going to use some cleaning paste and I know a lot of hospitals you may only have access to regular toothpaste you can use that but just keep in mind that regular toothpaste can be abrasive to the dentures if they have whiteners in them but we're going to use a cleaning pace and we're going to put it on our toothbrush just like how you normally would your own toothbrush at home and put the cap back on the toothpaste and set our brush down for a second while we get everything else prepped now we're ready to clean the dentures so what we're gonna do is I'm just going to rearrange the supplies just here a little bit that up here at our base in here and what we're gonna do is we're going to dawn our gloves because any time you're touching dentures you need to always wear gloves to protect you from getting any germs from the dentures because the mouth is has a lot of bacteria in it and we're going to turn our water on warm we don't want it hot because it can damage the dentures and we don't want it too cold because it really won't do the job so we're gonna turn our water on warm and we're gonna get our dentures and what we want to do first is we're going to rinse them off we're gonna put them in the kidney Basin because we want to clean this denture Cup so we'll have a nice clean area for the dentures to be returned to so grab your dentures and hold them firmly and then just rinse them hit him in the kidney basin now we're going to rinse our denture Cup and notice how dirty it gets you want to make sure you get all that out of there because we're going to fill it a little bit up with water just enough water to cover the dentures and why do we want to do that well we want to do that so the dentures are kept moist because it'll be easier for the patient to put them back in and it'll help them keep their shape now you always want to ask the patient what they use to soak their dentures in because some patients use half of a solution like half vinegar and half water or a tablet solution so always ask them that now what we want to do is we're going to start cleaning the dentures so we're going to get our toothbrush and we're going to start inward and work our way outward and see rinsing helps remove a little that but you have to watch out because down in the gumline you will get a lot of pace because a lot of patience use pace on their dentures and food particles and that's just a nasty reservoir for germs and bacteria to get in so making sure you really clean that area very good and then looking at the front of the teeth as well getting that cuz food particles like to get in there no one wants anything in their teeth whenever they put their teeth back in so you want them nice and clean and then flip them over and get up where the palate part would be as well okay and then we're going to rinse those let's get them nice and rinse we put them in your denture cup now we're going to clean our bottoms and again we're gonna follow the same way we did with the other ones clean and inside that gum line really good and make sure all the bacteria and food particles and everything's removed and brush on the back teeth and the front teeth molars just like how you would brush your natural teeth getting inside where the tongue would lay and then return to your denture cup which has the water in it and we'll put the lid back on the cup now what we're going to do is we're going to rinse our toothbrush and our basin get that cleaned get the toothbrush and really rub your finger your gloved finger over that toothbrush to get all the food particles out and the toothpaste out get it nice and clean and then I like to take a paper towel and just dry that Basin out that toothbrush dried as well and just put that back in there then turn your water off don't forget to pull the stopper and let the water drain out and then we'll take our towels once all the waters drained out put it in the linen and we're going to doff our gloves and perform perform hand hygiene and then when you're done don't forget to provide oral care to your patient and assist them with putting their dentures back in if they choose to okay so that wraps up how to clean dentures as a nurse thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	Walker_Nursing_Assistive_Devices_Mobility_NCLEX.txt	hey everyone is cereth register nurse sorry and calm and today we're going to talk about walkers and as always when you get done watching this youtube video you can access the free quiz i will test you on this assistive device so let's get started when you're studying walkers for your exams there are some things you want to remember you want to remember the proper fit what are those characteristics that the Walker actually fits your patient how to walk with the Walker so the gait and how the patient should get up and set down in a chair so first let's talk about proper fit how do you know that this Walker actually fits your patient well before a patient even uses a walker for the first time it has to be adjusted and on most Walker's they're adjusted down at the bottom you have to adjust each leg there's four of those legs so you've adjusted the Walker now what do you look for these are things you want to remember for your exams well have the patient hold onto the hand grips while they're standing with the Walker and you're going to look at that elbow there should be about a fifteen to thirty degree bend in the elbows also when the patient holds their arms down at their side their wrist crease should line up with those hand grips those two things really tell you that this Walker fits this patient the next thing you want to know about is the gait how does your patient actually walk with the Walker and this is something that you really want to pay attention to for exams so you want to make sure that you're paying attention to what's moving first is that the Walker is at the weak side or the strong side and how that order goes so before a patient actually uses a walker for those first couple times you want to make sure that you have applied a gait belt to their waist for safety also you want to stand on the patient's weak side in case they start to stumble or fall you want to be there and you want to tell the patient before they start ambulating with their Walker that they don't want to stare down at their feet while they're doing it even though that just seems like something natural you want to do that can actually mess them up and cause them to fall they want to look straight ahead just like if they were walking normally and the starting position how do they start out walking with their Walker well they want to make sure that the back tips of the Walker match up with the middle of their foot now let's talk about how you actually ambulate with a walker to ambulate with the Walker first the patient's going to get in position and hold on to the handgrips of the Walker first they will lift and move the Walker forward and then make sure all four points of the Walker are touching the ground then they will move the weakside put weight on the hands via the hand grips and then move the strong side again they will lift the Walker move it forward make sure all four points are on the ground then they will move the weak side put weight on the handgrips and then move the strong side you to sit down in a chair a patient is going to take their Walker hold on to the hand grips and slowly back up to the chair until they fill the chair with the back of their legs then they're going to slightly extend that weak leg out and take their hands and position them behind them and bin their strong leg and feel for the chairs armrests and then set down to get up from the chair what the patient's going to do is of course make sure the Walker is out in front of them they're going to lean forward out of the chair make sure their hands are on the hand rest and slightly extend that weak leg out then they are going to put weight on their hands by pushing up on the armrests of the chair and with their strong leg and putting their hands on the hand grip of the Walker then they are ready to ambulate and again to do that they will lift the Walker make sure all points are on the floor move that weak leg put weight on the handgrips and then move the strong side thank you so much for watching don't forget to take the free quiz and to subscribe to our channel for more videos
Nursing_Skills_Videos	3Point_Gait_Crutches_Walking_Pattern_Demonstration_Nursing_Skill.txt	hey everyone it's a registered nurse re and calm and today we're going to demonstrate how to do the three-point gait while using crutches and this is where they move both crutches and the injured leg together at the same time and then they will move the non injured leg thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	Pregnancy_Test_Positive_vs_Negative_Results_Time_Lapse.txt	here we have our droppers and they contain our urine sample we use the droppers that came with the pregnancy test to withdrawal the urine in this top dropper we have urine that contains HCG and how I know that is because it's my urine and I'm pregnant this bottom dropper contains urine that does not have HCG in it it comes from my husband so he shouldn't have HCG in his urine so what we're gonna do is we're gonna add three drops of urine to each test so add to this top one this is the urine that has the HCG so 1 2 3 and then wad drop here 1 2 3 and we're gonna let that go for 3 minutes cuz for this particular test they say to wait 3 minutes to read the results but don't read the results after 5 minutes because they won't be valid you so it's been three minutes now we're going to interpret our results so as you can see here in these tests we have a line in both of the control areas this tells us we added enough urine and that this test is working however in this pregnancy test here on the top there is a line in the test area so it's positive which tells us that it detected HCG in the urine however here on the bottom this bottom pregnancy test there is no line in this test area so it is negative it tells us that it did not detect HCG in that urine sample
Nursing_Skills_Videos	PPE_Gloves_Training_Don_Doff_Gloves_with_Personal_Protective_Equipment.txt	hey everyone it's sarah thread sterner sorry and calm and in this video I want to demonstrate how to Don and doff gloves whenever wearing PPE first reform hand hygiene then dawn gloves by pulling each glove over the hand and extending the cuff of the gloves over the sleeves of the gown next remove the gloves do not touch the outside of the gloves because they are considered contaminated to remove the gloves start by taking your non-dominant gloved hand to grab the other glove around the cuff in the wrist area do this by using a pinching motion to grab it peel this glove off by turning it inside out and what it into a ball with your gloved hand keep it securely in your gloved hand then take the index finger of the unloved hand and slide it carefully under the cuff of the gloved hand and peel the glove off the hand by pushing the index finger forward against the glove this will turn the glove inside out dispose of the gloves then perform hand hygiene okay so that wraps up this demonstration on how to Don and doff gloves and be sure to check out the other videos in this series
Nursing_Skills_Videos	How_to_Take_a_Pregnancy_Test_at_Home_Pregnancy_Test_Results_Live.txt	hey everyone its air thread sterner sorry and calm and today I want to show you how to take or administer a pregnancy test so what does this test measure well it's going to look for a hormone in your urine called HCG and HCG stands for human chorionic gonadotropin which is why we call it HCG and this is a hormone produced by the placenta so if this hormone is present in your urine it's possibly saying that during your cycle you release an egg a sperm fertilize the egg the egg traveled down and planted itself in the uterus and you are growing a baby so when should you take a pregnancy test well generally speaking you want to wait until you have missed your period by there are some tests on the market that say they can detect it a little bit earlier but always just read the instructions of whatever type of test you are using now with these tests they will display results differently depending on the brand some will display it electronically others you will have to manually read it like with this one and generally speaking as well whenever you want to take the pregnancy test is usually in the morning time whenever your urine is highly concentrated with the HCG what supplies will you need to take a pregnancy test well of course we'll need the pregnancy test you'll need a cup to catch your urine in you'll need a timer either a watch or your phone and if you're in a clinical setting you'll need a pair of gloves now to actually take the test what you want to do is you want to wash your hands then collect the urine specimen by peeing in a cup and then wash your hands again so we've collected a urine specimen and I've put on my gloves but before we actually do the test let me explain this test to you whenever you do a pregnancy test always look at the test itself because here this area where this arrows pointing is where we're going to drop the urine in and this particular test requires three drops of urine and then over here you see a C and a T the C is the control area and you will definitely get a line here and this is a good thing because this is telling you that the test works and that you added enough urine the tea part is the test part and if it is positive you do have HCG in your urine a line will pop up here now the darkness of the line how fast it pops up will vary from person to person depending on how far along you are in your pregnancy so read the manufacturer's results for how long you need to wait before you read the test results now we're going to do the test and what I'm going to do is I'm going to show you a test that is positive versus a test that is negative so you can see the differences here we have our droppers and they contain our urine sample we use the droppers that came with the pregnancy test to withdraw the urine in this top dropper we have urine that contains HCG and how I know that is because it's my urine and I'm pregnant this bottom dropper contains urine that does not have HCG in it it comes from my husband so he shouldn't have HCG in his urine so what we're gonna do is we're gonna add three drops of urine to each test so we'll add to this top one this is the urine that has the HCG so one two three and then we'll drop here one two three and we're gonna let that go for three minutes cuz for this particular test they say to wait three minutes to read the results but don't read the results after five minutes because they won't be valid you so it's been three minutes now we're going to interpret our results so as you can see here in these tests we have a line in both of the control areas this tells us we added enough urine and that this test is working however in this pregnancy test here on the top there is a line in the test area so it's positive which tells us that it detected HCG in the urine however here on the bottom this bottom pregnancy test there is no line in this test area so it is negative it tells us that it did not detect HCG in that urine sample so what do you do if you get positive results well what you want to do is call your doctor your OBGYN family doctor whoever you go to tell them that you got a positive home pregnancy test and they'll have you come in they'll confirm it and then start with prenatal labs but let's say that you get a negative result but you still haven't had your period you don't really know what's going on what you want to do is you want to wait a little bit and retest maybe you test it a little bit earlier or let's say it's still negative and you're not getting a positive result you want to call your doctor who can further investigate it okay so that wraps up this demonstration on how to take a pregnancy test now be sure to check out my other nursing skills videos and our maternity review lectures
Nursing_Skills_Videos	How_to_Use_a_Walker_How_to_Walk_Ambulate_with_a_Walker.txt	hey everyone in cereth registered nurse RN comm and in this video i'm going to demonstrate how to emulate with a walker to ambulate with the walker first the patient's going to get in position and hold onto the hand grips of the walker first they will lift and move the walker forward and then make sure all four points of the walker are touching the ground then they will move the weakside put weight on the hands via the hand grips and then move the strong side again they will lift the Walker move it forward make sure all four points are on the ground then they will move the weak side put weight on the handgrips and then move the strong side you thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	Chest_Tubes_Nursing_Care_Management_Assessment_NCLEX_Review_Drainage_System.txt	hey everyone it's Sarah with register nurse rn.com and in this video I'm going to be going over an inlex review about chest tubes what I'm going to do in this video is I'm going to cover the things you need to know exactly for the inlex exam I'm going to cover the anatomy the different types of Drainage Systems the purposes and the nursing interventions whenever you're maintaining these systems and after you watch this lecture be sure to go to my website register nurse aan.com and take the free quiz that will test you on your knowledge about how to take care of chest tubes and a card should be popping up so you can access that so let's get started first let's start out talking about the purpose of a chest tube what is it and what does it do okay it is a tube that is inserted into the plal space of the lungs to remove air or fluid to help re expand the lung so let's look at what it does okay we have our lung here that is that little red area on the drawing then around the lung we have a little protective layer called the visceral plora and then we have this small little space and then around that space that attaches to our thoracic cavity is called the paral plura and what happens is that whenever you breathe in and out these two layers Glide gently over each other because in the pluris space there's a small amount of Cirrus fluid that allows them to Glide nice and gently to prevent them from rubbing together but while they're gliding they're creating a negative pressure which allows your lungs to work properly and to inflate and deflate now whenever something gets into this space like air or fluid the lungs are like oh no this is not right it's messing up our pressure because we have a negative pressure and it causes the lung to collapse so the physician will go in there and insert a chest tube to help drain out that air and that fluid that's causing this lung issues now there's another type of chest tube which is um called a media styal chest tube and this is a tube that that is inserted into the medianum space and it's typically placed under the sternum to to drain fluid from around the heart after cardiac surgery because a lot of times after card cardiac surgery there can be extra blood or fluid and this can get around the heart and compress it and send the patient into cardiac tampeno so those are placed as well now let's take it a little bit deeper and look at some other reasons why a chest tube might be placed okay the first one we hit on this is a pneuma thorax and this is just where air enters into this small little space and causes the lungs to collapse this can happen because of like trauma or spontaneously happen I've had patients who've been admitted they the doctor wasn't sure why they had a numa thorax it just happened spontaneously another thing is called a plora effusion and there's different types of PL fusions depending on what type of fluid is in that PL space and what it is it's just fluid in the plor space so here's your nice little space and some fluid gets in there messes up that pressure are setting and the lung collapses and they have some major breathing issues okay different types of plor fusions you have hemothorax which is where blood enters into the plor space hemo means blood and this can be due to trauma a a disease like tuberculosis or a blood clotting issue they're um not clotting so blood is leaking in there another thing is called epema where they can get an infection in the plor space and last type of PL Fusion is a silo thorax which is where lymphatic fluid can get into the plora space and of course another reason for a chest tube is the cardiac surgery now let's look at the different types of chest tube Drainage Systems um whenever you get a job as a nurse be sure to familiarize yourself with the different types of chest tube Drainage Systems your facility offers and make sure you get a good in service on that because different places have different chest tube Drainage Systems okay here are your basic ones I'm going to be going over um for inlex purposes the wet suction and the dry suction and let me go over the basic setup of how a typical chest strain is set up and then we'll talk about the differences between the two okay so you have your little suction device the tube will go will be coming from the patient and this tube right here is from the patient and it's draining down into the drainage chamber these are your drainage chamber where whatever is coming out of that lung is Flowing down into there then in the middle you have your water seill chamber and there's blue water in this and as the patient breathes in and breathes out this water will TI will osculate up and down and sometimes there's a little ball in there that will move as well and then you have a little mo an air leak monitor area and in this area you were looking for bubbling because if you see continuous bubbling which we'll go over in depth whenever I'm covering the nursing interventions and there could be an air Le and then over here at the very end you have your suction control chamber now notice on the wet suction and the dry suction it looks a little bit differently and that is the biggest difference with these two systems is the suction how the suction works so let's cover it okay wet let's talk about the wet section okay the wet section is regulated by the height of the water in the suction control chamber when it's connected to wall suction so whenever you're setting up a wet suction chest Tu drainage you will be filling this with the water that it comes with and um depending on what the physician orders you'll fill it up to whatever they order um typically it's -20 cm of water and this right here once you connect this tube to the wall section will regulate the suction control of the chest tube and you will hear bubbling and see gentle bubbling in this as it's working so that is normal now with dry suction the water there is no water column and the suction is controlled and uses a suction monitor Bellow that balances the wall suction and um you can adjust the wall suction pressure by using a little rotary suction dial on the side of the system so this area right here this is where your suction monitor Bellow is and it looks like a little orange accordion area and whenever you turn on the wall section to this tube this little orange accordion will start to expand out and you have this little triangle there that tells you once it gets to that triangle it's regulating suction it's good and then here you have your suction control regulator and you have a little dial on the side where you can set the prescribed suction of whatever the physician orders and here it's set on -20 like how it would be over there in the water suction now with the dry suction systems you can get a high you have higher suction pressure options there's no bubbling of water because you don't have a water column like how you do on wet suction and there's no water evaporation with the wet suction because you have this water you have to fill it over time this can evaporate so you'll have to monitor that make sure it's at a good level here you don't have that so you won't have water evaporation now let's look at our nursing interventions of things that you have to do for this patient who has a chest tube the biggest thing you want to do whenever taking care of a patient with a chest tube is you want to monitor the patient's respiratory status very closely you want to monitor the drain system itself and you want to know what to do when things go wrong like if the chest tube becomes dislodged accidentally or the system breaks and how to assist the physician with removing the chest tube and I'm going to be covering all those things so first let's talk about the drainage system and the tubing okay the drainage system itself needs to keep needs to be kept below the patient's chest and the tubing especially the tubing coming from the patient it tends to be long and bulky and patients roll over on it gets caught up in a side rail so you want to make sure that those connections are secured and that they're draining down into the system and that there's nothing no stagnant fluid collecting in those and clots and that your connections are sealed next while you are taking care of this patient with the chest tube you're going to be be watching The Collection chamber the water seal chamber and the suction control chamber and this is going to tell you a lot about what's going on with the patient but first let's talk about the drainage collection chamber the drainage collection chamber is whenever you're monitoring this you want to note the color of the drainage how much they're putting out typically less than 100 cc's per hour and you want to record it very very well regularly because Physicians are going to ask you how much is that chest tube putting out next the water seal chamber this is your water seal chamber on the dry suction and on the wet wet section what does the water seal chamber do it performs an underwater seal to allow air to be remove from the plor space while preventing outside air from entering into the lungs because remember we want to create a negative pressure in there because that's what the lungs like so that water still helps us do that now one thing you want to know this is normal the water will fluctuate in this water seal chamber it will osculate up and down so that's normal you want that and whenever the patient breathes in it the water height will increase and when they expire have expiration decrease the water will decrease however it's the opposite if the patient is on positive pressure mechanical ventilation whenever the vent breathes in for them the water will decrease when the vent breathes out for them the water will increase so just commit that to memory now inlex question what if the water in the water seal chamber you notice it's not fluctuating at all what could be the issue well the lung may have re-expanded corrected our problem or there's a kink somewhere so you want to check that out next the air leak monitor area this is part of the water SE chamber and it's at the bottom and what we're looking for in that is bubbling inlex loves ask questions about bubbling so what's the big thing normally there should be no bubbling in there because it's monitoring for heirs however if you have excessive bubbling noted in that area that could mean an air leak however if the patient has a numo thorax and there could be intermittent bubbling in this now let's think back to to a pneumothorax what is a pneumothorax remember we talked about at the beginning of the lecture it's where air enters in to the plora space so as that air as that patient is recovering air will escape and leak from the lungs into the water seal chamber so you could see a little bit intermittent intermittent bubbling so that could be normal for them but the excessive continuous bubbling is not that can indicate an air leak okay next part of it is the suction control chamber remember on the wet section we have the water column and on the dry section we have the suction Bellow um little regulator that works with that biggest thing you need to know is that um with wet suction you're going to hear a continuous bubbling noise and you're going to see gentle bubbling in this that is normal because it's connected to wall section and that's telling us it's working with and the water can evaporate over time so you want to make sure that you're adding water if it does evaporate evaporate to keep it at the prescribed amount of suction with the dry suction there's no water column and it's regulated by that suction monitor Bellow that little orange accordion thing other thing you want to do is you want to monitor your patient's lung sounds how fast they're breathing if they're having any complaints of difficulty breathing like dmia um watch the insertion side look at it make sure it's free from infection and also check for any subq crepus or subq osine also called that this is where carbon dioxide escapes into the tissues and whenever you feel it'll be puffy and you feel it you'll never forget it if you ever feel it it feels like a crackling sensation underneath the skin and also you're going to be having the patient cough and deep breathe that helps move fluid and keeps their lungs nice and functioning and you're going to be repositioning them okay what to do if the chest tube becomes dis loded if this happens cover the site with a sterile dressing and tape it on three sides doing this will allow air to escape and prevent a tension Numa thorax and notify the physician immediately okay what happens if the system breaks you walk into the room it's fell over cracked you need to get a new one while you're waiting on your new one to arrive order a new one um you'll take the tubing and insert it one inch into sterile water to make that water seal and get a new system okay what about milking or stripping the tubing this used to be done a long time ago however it's not really recommended anymore um due to increasing pressure so always follow your hospital protocols with this another thing clamping another issue always follow your hospital protocols what do they say to do with that um because there's an increased risk of increasing the patient's chances of developing tension numo thorax and never do it without a physician's order okay so removal of a chest tube typically the Physicians will do this in some facilities nurses have been checked off through competencies to do this but typically for inlex purposes you will be assisting The Physician usually done at the bedside and what you will do is you will gather the supplies uh typically varies on physician preference so always make sure you know what your Physicians like uh sterile gloves dressing supplies um this could be a clusive petroleum base tefla whatever a mask gloves a suture removal kit and rubber tipped hemostats okay one thing you're going to be doing prior prior to removal you're going to teach the patient how to do what's called the Val Salva maneuver and this is where you will have the patient take a deep breath exhale and bear down and they will do this during removal the reasoning for this is to prevent air from entering that plural space during removal so that helps decrease that from happening then if ordered uh you'll premedicate the patient for pain because this can be painful uh position the patient in semi fowers position and afterwards you're going to monitor the respiratory status listen to those lung sounds watch for equal chest rise and fall make sure it's not unequal any drainage is the patient breathing okay are they complaining it's hard to breathe and typically after a removal is done the physician will order a chest x-ray to assess lung expansion so that is an inlex review about chest tubes now go to my website register nurse rn.com and take the free quiz to test your knowledge and be sure to check out my other inlex review videos and thank you so much for watching and please consider subscribing to this YouTube channel
Nursing_Skills_Videos	9_Pulse_Points_Assessment_on_the_Body_Nursing_Anatomy_and_Physiology.txt	hey everyone in cereth register nurse re and calm and today i'm going to demonstrate how to find the 9 most common pulse points as the nurse it's important to know these common pulse sites because some of them you will be using routinely during your assessments now the ones that you're not using routinely it's still really important to know where they are located and whenever you're assessing the pulse you will be looking at a few things one thing will be the rate how fast is it along with the strength and you'll be grading it on a scale zero to three with zero being absent one plus it's week two plus it's normal and three plus its bounding and then you'll want to look at the rhythm is it regular or irregular and as you fill on the pulse some of the sites you will be filling bilaterally to see if they're equal the only one you really don't feel on at the same time is the carotid we will be filling on that one at a time now the nine pulse points we will be assessing in this video will be the timbrel the carotid apical brachial radial femoral popliteal posterior tibial and dorsalis pedis to find the pulse points we're going to start from her head and work our way down it just makes it easier going in that order and to find the pulse points you can use your first two or first three fingers and we're going to find the temporal artery and to find this artery you'll want to find landmarks and this always goes for any pulse point that you're trying to find now this temporal artery comes off of the external carotid and goes up and what you want to do is you want to find the tragus of the ear which is this part of the ear and the zygomatic arch is found above it which is a fancy way of saying the cheekbone so your pulse site where this X is is found right here this is your temporal artery and you'll want to feel bilaterally and see if they are equal and the time that a nurse really feels on this is during that head-to-toe assessment when whenever we're assessing the pulses in the head the next pulse point we're going to assess is the carotid artery and it is found in the neck and this artery is most commonly assessed during CPR in an adult and it supplies our brain and our head with blood now whenever you are assessing the carotid artery as I pointed out at the beginning of the video you will assess each side at a time you will not do it bilaterally because we don't want to stimulate the vagus nerve which can drop the heart rate and decrease circulation to the brain to find this pulse point we'll use the landmarks of the jaw because we're gonna go below the jaw we're gonna have the patient tilt the head like this and we're gonna find the trachea and the sternocleidomastoid ahead to toe assessment or before the administration of digoxin and in an adult we want to make sure we listen to the apical pulse with our stethoscope for one full minute and to hold the medication if apical pulse is less than 60 beats per minute and this side is the point of maximal impulse and is located at the apex of the heart so it's going to be on the left side of the chest at the fifth intercostal space midclavicular lee so what you want to do it's best if you have your patient lie on their back and you'll need your stethoscope by the way because you want to actually listen to the pulse and count it so how you do that you want to find your landmarks so you want to find your sternal notch which is this notch up here then go down the angle of Luis and you're gonna go to the second intercostal space midclavicular so this is about mid clavicular we're gonna go to the second intercostal space which is about right here now we got to get to the fifth intercostal space because this is where our pulse is so we're gonna go three four five so our pulse is located here in this area then we're going to take our stethoscope and we will listen and assess the pulse now it's fine PulsePoint in the arm which is the brachial artery and this is a major artery found in the upper arm and it will actually go and divide into the radial and the ulnar artery now we use this pulse point for whenever were assessing the blood pressure and during CPR in an infant we will check this site and how you want to check the size you want to have the patient extend their arm and turn their arm out where their palms are facing upward to find this artery you want to find the area of the cubital fossa which is this triangular area in front of the elbow and the brachial artery is found near the top of this cubital artery in this location right here next we're going to find the radial pulse point and this artery branches off from the brachial artery and helps provide circulation to our arm and hand and we most commonly use this pulse point for measuring a pulse rate in an adult and to find that how you're going to do that is just extend the arm out and make sure the palms are facing upward just like how you did with the brachial artery this artery is found right below the thumb within the wrist area along the radial bone and you will suss it in this area right here next we're going to find the femoral pulse point and this is a major artery in the groin that provides circulation to the legs and to fill this artery you have to palpate deeply in the groin area and it is found below the inguinal ligament and between the pubic symphysis and the anterior superior iliac spine next we're going to find the popliteal pulse site and this comes off the femoral artery and it's located behind the knee in these areas right here to find this pulse point you're going to flex the knee and you're gonna take both your hands put them behind the knee and you will find it at about the middle area of the popliteal fossa which is a diamond shape pitted area behind the knee and this artery is pretty deep so helping to the knee will help you find that artery now we're going to find the posterior tibial and the dorsalis pedis and these sides are assessed during your head-to-toe assessment especially if your patient might have peripheral vascular disease or if they've had a vascular procedure like a heart cath where they access the femoral artery we'll always be checking the pulses in these lower extremities now the dorsalis pedis want to turn the fit a little bit this way is found here and in order to find that what you can do is find the e HL tendon the extensor helices longus tendon that actually helps extend this big toe so if you've lift your toe it for me you can see that tendon dry in here now just follow that tendon and then just go a little bit to the end of it and that will be your dorsalis pedis then we'll find the posterior tibial and this is found on the inside of the ankle found between the back of the medial malleolus which is the bony prominence of the ankle bone and Achilles tendon so you'll find it within this area right here okay so that wraps up this video over the nine most common pulse points and remember whenever you are counting the pulse rate if it's regular you will count it for 30 seconds and multiply it by two if it's irregular you will count it for one full minute and we always count the apical pulse for one full minute thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	Pregnancy_Urinalysis_Protein_Glucose_in_Urine_Reagent_Test_Strips.txt	hey everyone is cereth register nurse sorry and calm and in this video i'm going to demonstrate how to check the urine for glucose and protein which is done during prenatal visits so why don't we check the urine of a pregnant woman for glucose and proteins well pregnant women are at risk for gestational diabetes and preeclampsia and I have some videos over those topics if you would like to review those but we're looking for glucose in the urine because this could indicate gestational diabetes and we're looking for protein which could indicate that the woman may be developing preeclampsia so you'll want to of course perform hand hygiene dawn some gloves you'll need the specimen provided from the pregnant woman you'll also need a test strip and you'll need the diagram because you'll be referencing this test strip with these colors so I've performed hand hygiene and I'm donned gloves now we're going to take this test strip and we are simply just going to dip it in the urine like this and then we are going to take it and lay it like this and we're gonna wait for one minute and read results so it's been a minute now we're gonna take our test strip and we're going to compare it with the little color key the top one is the glucose and based on our color it is normal for glucose the bottom one is checking for protein and based on its color it is negative for protein after you interpret the results you want to dispose of the urine per your facility's protocol Dolf your gloves and perform hand hygiene okay so that wraps up this video on how to check the yearn for protein and glucose and be sure to also check out our other videos that demonstrate nursing skills and the maternity reviews
Nursing_Skills_Videos	How_to_Wear_Don_Take_Off_Doff_Surgical_Face_Mask_Tutorial_PPE.txt	hey everyone it's sarah thread sterner sorry and calm and in this video i'm gonna demonstrate how to wear and take off a mask first let's talk about some of the common mistakes that people make whenever wearing a mask like this one common mistake that people make is that when they wear the mask they will wear it under the nose another common mistake is that people will fail to seal the top of the mask around the nose and cheeks in addition people will sometimes fail to pull the mask under the chin people also wear the mask upside down or backwards and finally while wearing the mask people will sometimes touch the front of the mask which is considered contaminated now let me give you some tips on how to wear the mask so first you need to determine which part actually goes over your nose this part or that part so to do that what you want to do is you want to fill along those edges and where your feeling for it is like a metal type band that's sewn into the mask that's easily flexible that can fit over your nose and that's the top part next what you want to do is you want to determine which part of the mask is the front versus the back and there's two ways you can do this and if you're using a mask with ear loops you can look at the ear loops and where they are sewn at this will be the area that goes on the face so on this mask they're sewn here so this is the part that goes over my face also you can look at the pleats of the mask and look at how they are pointing when the pleats are pointing downward this is the part that goes on the outside versus the pleats that are pointing upward that's the part that's going to touch your face before we put on a mask we want to perform hand hygiene when using a face mask with ear loops place each loop over the ears then grasp the nose piece of the mask and bring it to cover the bridge of the nose then mold the nose piece of the face mask with the finger tips of both hands by starting at the bridge of the nose and work outward toward the cheekbone then grasp the nose piece of the face mask and pull the bottom of the mask under the chin when removing the mask it's important to remember that the front of the mask is considered contaminated therefore grasp the ear loops of the mask with the fingertip and removed from the face dispose or clean and store for future use the mass / facilities protocol and then perform hand hygiene ok so that wraps up this video on how to wear and take off a mask and don't forget to check out the other videos in this nursing skills series
Nursing_Skills_Videos	Sterile_Gloving_Nursing_Technique_Steps_DonDonning_Sterile_Gloves_Tips.txt	hey everyone is sayers read sooner sorry and home and in this video I'm going to show you how to dawn sterile gloves the very first thing we want to do is wash your hands then select a pair of gloves that will fit you because these come in various sizes once you have your gloves you will open them up like that and you will notice these little flaps these flaps will assist you and opening up the glove so you won't contaminate them so just put your fingers underneath them and just pull it open and remember you have two inches to grab around this field and you do not want to cross two inches because you will contaminate the field so just go around and just gently open the gloves up and you have your right glove and your last glove and these are the cuffs which we will be touching and they will become unfair oh but we do not want to touch the outer part of the glove so first what we're going to do is we're going to glove or dominant hand and I my dominant hand is my right hand so I'm going to glove that first so what I want to do is I'm going to take my left hand and I'm going to grab the cuff of the right part of the glove and I'm going to slide my hand into the glove and I'm going to slightly tuck my son-in so it'll be easier getting the glove on so I got the cuff I'm going to slide my hand in and then just put my fingers out and then I'm just going to roll that down the cuff making sure I don't touch the outer part just so it slides up my arm okay now this is sterile do not touch anything other than what you're going to be doing with this glove because it will become contaminated now what I'm going to do is I'm going to just simply take my fingers underneath the last glove and get under there so I can glove my left hand so just sliding underneath now I'm going to make sure I keep this thumb away because I'm going to slide it over my left hand and I'm going to slightly just tuck my left thumb in again and just slide underneath the glove and I'm pushing with this part of the glove so it will go over my hand and have that and then I'm just going to just gently roll it down without touching my arm and I have the gloves on and they are sterile so now I'm ready to perform my procedure thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	Pressure_Ulcers_Injuries_Stages_Prevention_Assessment_Stage_1_2_3_4_Unstageable_NCLEX.txt	this is cereth registered nurse re and calm and in this video I want to be going over pressure injuries formerly called pressure ulcers in this video is part of an Inc Lex review series over the integumentary system and as always I've been to this YouTube video you can access the free quiz that will test you on this condition so let's get started pressure injuries have had various names over the last several years and the past we have called them pressure ulcers decubitus ulcers or bed sores and right now they're most commonly termed as pressure injuries so what are they they are the breakdown of skin integrity due to some type of unrelieved pressure and this can be from a bony area on the body coming into contact with an external surface which leads to a pressure ulcer and here in a moment we're going to go over all those common sites of those bony sites that can lead to a pressure injury also some type of medical device can cause this or friction and shear which we're going to talk about in depth here in a moment so now let's talk about how a pressure injury happens okay to do that we first need to review the basic layers of the skin so we have the epidermis on the top then below that we have the dermis and then right below the dermis we have the capillary bed which feeds and perfuses and supplies our dermis and epidermis so remember that and then below that we have our fatty subcutaneous tissue followed by muscle and then bone okay so we have a patient and let's say they've been sitting in the chair for a really long time okay so when a patient studying in the chair what bony prominence is going to put them at risk for developing a pressure injury that coccyx bone so what happens say they've been in the chair way too long that coccyx bone is exerting pressure this way on those bottom layers then you have the external hard surface that chair with a patient setting and it's exerting its pressure down this way now what's in the middle we have pressure coming this way and that way we have our blood supply which have two pressures coming together two layers pushing together it's going to pinch off that blood supply to that skin the epidermis and dermis which is going to lead to what a pressure injury so we have various stages of pressure injury it goes from stage one to stage four and we're going to go over those here in a moment I'm going to give you examples and whenever we are looking at pressure injuries we're looking at the depth of how much this injury has been affected for instance does the injury extend down into where you can see the subcutaneous tissue which would be like a stage three or does it even go as far where you can actually see the muscle the tendon the ligament and the bone which is a stage four or it can be where the top layer of the skin is completely intact but it's really red and when you press on it it doesn't blanch so it doesn't turn white and that would be a stage one now let's talk about friction and shear for a moment okay what is this because as a nurse when we get into our nursing interventions we really want to make sure we're not doing activities that caused this because that can cause a pressure injury so what this is it's where you have the bone right here the bottom layer moving in the opposite direction of the skin so how can this happen big common way this can happen is if your patients sitting in bed and they're sliding down in the bed so what happens is that we'll say the coccyx again is moving this way and as they slide down the pit the skin is moving this way so when you have these two forces going opposite direction what's going to happen to that middle layer that's so important that perfuses our dermis and epidermis it's going to tear it's going to mess up then we get decreased perfusion which makes a great site for a pressure injury now let's talk about risk factors that can increase a patient's risk of developing a pressure injury and I would really remit these risk factors because exams love to throw out all these patients that could have a potential risk for a pressure injury and you need to select the one that has the most risk okay so whenever you're looking at those options or you're thinking about this try to think about it this way okay think of the patient population that can't relieve their own self pressure like they can't shift their weight in bed or they have decreased sensory perception which would lead to them not really being able to shift their own way or they've had some type of injury that prevents them from doing that have some type of medical device they're wearing like the splint the nasal cannula really with the nasal cannula I have seen where they have wore pressure injuries on the ear so you have to watch out for that or they have decreased skin integrity so poor nutrition is definitely a huge risk for developing a pressure injury so look out for those low weights low body mass indexes because I decreases our skin integrity if the patient is in mobile they're just not getting out of bed they can't move they can't get up and walk they're confined to that bed which requires you to turn them every two hours definitely at risk they have some type of neuro issue where they have decreased level of consciousness this could be because they're sedated we're in a sense causing their decreased level of consciousness because we're treating them for something else or they have their labs are thrown off or they're just not aware of where they're at or they have some type of spinal injury maybe in quadriplegic something like that where they can't actually feel that pressure and shifts their own weight diabetics are at risk for this so if they're in mobile have neuro issues poor nutrition really huge candidate for a pressure ulcer because why is that well diabetics we know that they have decreased perception of sensory so patients who don't have diabetes they can feel you know if they step on something with their feet more than a person who has diabetes they may not know that they've injured their foot and they have decreased perfusion so their epidermis and dermis isn't going to be as perfused as well as one who doesn't have diabetes okay so who is incontinent this can be a stool or urine if that sets on the skin long they have that pressure building that increases the risk of a pressure injury and activities that cause that friction and shear that we just talked about so preventing your patient from sliding down in the bed and when you do have to move your patient up in bed because a lot of times patients are they get down in the bed instead of dragging them up in the bed lifting them and moving them instead of making it where you would drag them because again the bones gonna move in another direction the top layer of skin is going to move on the other in a tarry capillary bed and increase the risk of a pressure injury now let's talk about the top sites for pressure injury and whenever you are a nurse taking care of your patient you want to be aware of what position the patient's in and where those bony prominences are at so you can be thinking of okay I want to make sure that they're not having pressure injuries at these locations and these locations include the heels and the ankles as well as the hips the sacral area on the back the elbows and shoulders the inside of the knees along with the occipital area and the ears now is like a little quiz okay we have this patient they're laying on their left side so where do you think they're gonna be at most risk for a pressure injury so whenever we look at this we need to think okay what areas are going to be either coming into contact with the hard service that bony prominence or are they going to be too bony prominences coming together because I can also lead to a pressure injury as well okay well suffer start from head to toe and work our way down so the ear definitely because laying right there that's a perfect area for a pressure injury next move down our shoulder hard bony area on your shoulder right there can lead to one and then as we move down our hips big bony area that's just a perfect site for a pressure in jury the knees whenever you're laying in this position not only with the knee coming into contact with the surface but the two knees actually coming together can cause that along with the ankles which is another side so the ankles coming together or the ankle lane on the bed so it would be good to get this patient like a wedge pillow to help prevent that from happening now let's talk about staging of pressure injuries and these stages are based on the national pressure injury staging system okay Stage one this is where the skin is completely intact so that top layer is not going to be broken and this area is going to be very red but the key would say to one that you need to remember is that it doesn't blanch so if you touched on this red area it would not turn white it would stay the same color and here is an example of a stage one if you pressed here on the hill where it's extremely red it would not blanch next h2 this is where the skin is visibly damaged it's not going to be intact and you are going to see partial loss of the dermis so extends down to the dermis but it will not extend down into the sub-q fatty tissue that will not be visible and it can have a wound that can be a superficial red or pink open ulcer or it may have a formation of an open or closed blister and here's an example of what a stage two looks like next is stage three this is where the skin is visibly damaged and not intact with full loss of skin tissue now you may see the sub-q fatty tissue which will be yellowish or white in that wound bed and the wound edges can be rolled away like called epi ball now the big thing about this is that you are not going to see tendon muscle ligament that will not be visible and here is an example of a stage 3 and notice at the top that the wound edges have began to roll away and one of the most severe stages of a pressure injury is a stage 4 this is where the skin is visibly damaged of course and there is also full loss of skin tissue but the thing is is that it will expose bone muscle tendon and ligaments you will be able to see this and here is an example of a stage 4 pressure injury and as you can see you can see the bone very clearly in this image now let's look at an unstageable pressure injury ok the reason it's unstageable you can't actually stage it as a 1 2 3 or 4 is because there slough or eschar covering a full thickness ulcer and Slough is yellowish 10 substance and you can see that here in this picture and eschar is like a brownish black area covering the womb so because that's in the way you can't actually see the depth of the wound so it's labeled unstageable and lastly let's look at a deep tissue injury ok this presents as a purplish or blackish area over the skin nice intact and the fatty tissue is injured below and you may also see where there's been like this blackish full looking blistered area I've seen this on the hills and when you touch this palpate it it will feel heavy and spongy and this is a deep tissue injury now let's talk about nursing interventions what are we gonna do for this patient who has a pressure injury ok our role includes prevention detection and wound care so what we're going to be doing is whenever we receive a patient whenever we go in to do that head-to-toe assessment we are going to really concentrate on that skin integrity and making sure that they have no pressure injuries already there and if they do we immediately want to document that and we want to include the stage or the pressure injury the size the color the drainage also notify the physician if it's a really bad wound that can be cultured you want to notify them about that so you can possibly obtain a order for a wound culture also if the wound is very severe you may need an order to contact the wound care team so they can come assess that wound and prescribe treatment another thing is every shift you're going to be assessing the patient's risk factors for a potential pressure injury and in the hospital with every shift when we document we use the Braden scale and the Braden scale looks at six categories it looks at sensory moisture activity mobility nutrition and friction and share which is what we really talked about at the beginning of the lecture with those risk factors so we're looking at all those and based on what they score it can go from zero to 23 with a nine or less meaning a very high risk in 19 to 23 there's really no risk at all now in our plan of care what we want to include are things that will help us prevent any further breakdown of a pressure injury that is already there or to help prevent one altogether so some things we can include as keeping the skin dry and clean the patients incontinent you want or excessively sweaty you want to use barrier creams to help protect that skin especially where pressure injury can develop make sure they always are wearing a clean gown dry linens and preferably wrinkle free also turn them every two hours that's the minimum we want to make sure that if they can't alleviate their own pressure on those bony prominences that we're helping them do that watch for those activities that produce friction and shear and we talked about that earlier in the leg and one thing with the patients who really slide down in the bed what can help is as you put the head of the bed up put their foot of the bed up a little bit so they're not gonna just be sliding down and said when you have the foot up a little bit that will help prevent them sliding down and another thing is that you can get your patient special devices like there are air beds made for patients who are at risk for pressure injuries if they are just at a huge risk they score really low on the Braden scale you can get them beds that will cycle on and off with these air pockets that will help alleviate the pressure on those bony prominences also you can get patients Hill boots to help prevent those hills from just setting on the bed elbow pads as well as wedges to keep those knees and ankles separated gel cushions for the bottom if they set in the bedside chair a lot or they set up in bed or they're in the wheelchair that can help prevent pressure injuries developing there some other things include to routinely assess the skin integrity around those medical devices especially with the nasal canula any type of mask they're wearing or if they have splints like our ortho devices it can really put pressure on the skin and lead to a pressure injury so make sure you're always assessing that also the physician may order to consult nutrition or you can recommend that especially if your patient has poor nutrition and it's going to alter the way that they're wound is going to heal so they can prescribe high caloric diets with all those nutrients needed to promote wound healing also a lot of times whenever the pressure injury is very severe they will console wound care which will come in look at the wound prescribed treatment and as the nurse you will need to follow that out based on their regimen that they lift them and there's various treatments for pressure injuries that depends on the stage and the severity and what's going on some things include wound vacs debridement special dressings that will promote the wound you in that womb base and a neat treatment called hyperbaric oxygen therapy and this is where they deliver high amounts of oxygen to that wound to help promote healing okay so that wraps up this review over pressure injuries thank you so much for watching don't forget to take the free quiz and to subscribe to our channel for more videos
Nursing_Skills_Videos	How_to_Go_Up_and_Down_Stairs_on_Crutches_Nursing_NCLEX_Review.txt	hey everyone it's sarah thread sterner sorry and calm and today we're going to demonstrate how to go up and down the stairs while using crutches how does a patient navigate up and down the stairs while using crutches well you want to keep these two straight and what I'm meaning is which leg is going to go first up the step versus which leg is going to go down the step first and to remember that remember good up and bad down so whatever a patient is going up the stairs their good leg is going to go first up on the step followed by the crutches and the bad leg which will proceed and go up the step now whenever they're going down the steps they're going to move the crutches down onto the stab that will help provide stability followed by the bad leg because a bad leg is going to go down and then they're going to move the good leg down on to the step thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	Infant_Vital_Signs_Pediatric_Nursing_Assessment_Newborn_NCLEX_Review.txt	hey everyone it's Sarah with registered nurse rn.com and in this video I want to demonstrate how to assess vital signs on an infant so let's get started when assessing vital signs on an infant you want to start with the most non-invasive Vital Signs first while the infant is calm and resting therefore first you're going to check the respirations followed by the heart rate then the temperature the weight the length the head circumference and then the chest circumference before you assess the infant's vital signs you want to make sure that you gather your supplies so what you're going to need is an infant size stethoscope followed by a measuring tape a thermometer a scale and a watch to count with and you want to make sure that you sanitize everything that you're going to use on that infant before and after use and of course perform hand hygiene before beginning first we're going to check respirations and we're going to do that first while the infant is laying here nice and calm so we get an accurate reading and a normal respiratory rate in an infant is about 30 to 60 breaths per minute so before we even start counting respirations we want to look at that infant and we want to make sure that they're not in any respiratory distress so one way we can tell is we can look at the nose infants love to breathe through their nose so we were checking their nostrils and we're looking for nasal flaring and if you see that that's not a good sign also you want to look at that chest and you want to look for anything like chest retractions and what that looks like it literally looks like the skin has just been pulled over the ribs and you can see the ribs very well and this infant does not have either of those things in addition you want to look at the infant's skin color and make sure it's appropriate for their skin tone and that you don't see any signs and symptoms of cyanosis so to count respirations what we're going to do is we're going to count for one full minute the rise in the fall of the chest Now with an adult you probably remember that if the rate was regular we could count for 30 seconds and multiply by two well this isn't the case with infant you're going to do it for one full minute because infant's breathing pattern is irregular they have what's called periodic breathing where they'll breathe and then they'll stop for a second so we're going to count for one full minute and we're looking at that rise and that fall of the chest and infants like to breathe abdominally so if you're having trouble what you could do is you could lightly place your hand on the chest and feel the rise in the fall but you want to be careful not to disturb the infant and get them crying so I'll demonstrate and I'll count with each rise and with each fall just to give you an example so here we go one two three four five six seven eight nine ten so you want to count that rise in that fall for one full minute and then document that number so his respirations were 38 which again is normal because a normal rate is about 30 to 60 breaths per minute to count the heart rate you want to make sure that you're using the appropriate size diaphragm and bell and we have an infant here so we're using an infant size and for an infant who is less than a month a normal heart rate is about 100 to 190 beats per minute and this can vary if the infant is crying or if they're sleeping now if the infant is over one month that heart rate goes down a little bit the normal ranges it's about 90 to 180 beats per minute so to count the heart rate what we're going to do is we're going to use the apical pulse we're not going to use the radial pulse like how we do in adults so to find the apical pulse what you want to do is you want to find the fourth intercostal space now on an adult remember it was the fifth fossil space at the microvicular line but for an infant it's at the fourth intercostal space lateral to the midclavicular line so you want to find the clavicle and then we have the breast bone and we're going to go down and we're going to fill in between those little ribs and we're going to go to the fourth space so there's one two three four and then we're going to just go lateral to the midclavicular line so we're going to go about right here and this is about where the apical pulses on our infant and we're going to place our stethoscope there and we're going to listen for one full minute now an infant's heart rate can have sinus arrhythmia where it's irregular because their heart rate is can be affected with the respirations where their heart rate will actually speed up and slow down with respiration so that's where you want to count for one full minute now counting an infant's heart rate can be difficult because it is fast and one thing that I have found that definitely helps me is I find that heart rate and then I like tap my finger along with that heart rate and count and our infant's heart rate here is about 142. to check the temperature in the infant we're going to check it via the axillary route so we're going to place the thermometer tip in the armpit of the infant and I've already placed a little protective sheath over the thermometer so a normal temperature in an infant can vary anywhere between 97.5 to 99.3 degrees Fahrenheit to take the temperature make sure that you turn the thermometer on and then you're going to place the tip of the thermometer deep inside the fold of the armpit and then you're just going to put the arm down and you're going to wait for the thermometer to beep to tell you the reading okay our thermometer is beeping so it is done and the reading is 99.1 so we would document this and this Falls within normal range to obtain a weight on an infant you want to remove the infant's clothing and any soiled diaper a dry diaper can be used so place the infant on the scale and obtain the weight here this infant weighs 10 pounds 2 ounces so you'd want to document that way and then look at the infant's previous weights this patient weighed 8 pounds at Birth and now at this two-week appointment they weigh 10 pounds 2 ounces so that is a very good weight gain from their birth weight to measure length what you want to do is you want to lay the baby on a surface that you can Mark you'll need a PIN for this and you may need someone's help to help you measure the baby because you're going to measure from head to heel so lay the baby back you'll want to put their head midline like this and then just mark their head and then you want to take the leg and extend it outward like this make sure it's nice and extended you may have to work with them a little bit and then Mark it at the heel and a normal length in an infant is about 18 to 22 inches and then we're going to gently lift the baby up so we're going to measure this and we're actually measuring this in inches and the infant is 22 inches to measure head circumference you're going to need a measuring tape and you want one that measures in centimeters now a normal head circumference in an infant is about 33 to 38 centimeters and what we want to do is we want to place the tape measure around the largest diameter of the head so to do that we're going to place a little bit above the eyebrows and then we're going to wrap that around and place it around the most prominent part at the back of the head so let's do that so gently just lift the head up and put the measuring tape behind it and then pull your measuring tape around remove his head a little bit and let's see what we get and his head circumference is about 37 centimeters to measure chest circumference you want to get your measuring tape again you'll be measuring this in centimeters and you're going to take it and use the nipple line as your guide so you're going to wrap it around there and then measure it and his is about 36 and this should be about one to two centimeters less than what the head circumference was okay so that wraps up this demonstration on how to assess vital signs on an infant and be sure to check out the other videos in this Pediatrics series
Nursing_Skills_Videos	Counting_Respirations_Nursing_Skill_Assessment_Respiratory_Rate_CNA_Skill.txt	hey everyone at Sara thread sterner sorry ENCOM and in this video I want to give you some tips on how to count a respiratory rate so Kelly our respiratory rate seems like a pretty simple skill but it can be tricky especially if you're first starting out and you're trying to collect all those other vital signs on top of your respiratory rate plus some patients it's easy to count the respiratory rate while in others it's not first let's quickly review how to count a respiratory rate so a normal respiratory rate in an adult is about 12 to 20 breaths per minute but not only are we counting that rate we're looking to see if the breathing is labored or unlaid an example of labored breathing would be like this and notice how the patient's chest we can really see it moving up and down they're using their mouth to breathe in and breathe out also we're looking at if the breathing is irregular or regular so irregular breathing would be patients taking a breath here and there it's not on this consistent rate they may like take a breath in and out in and out and then there's a pause and then they'll breathe again here and there a regular rate would be there's a consistent rhythm to that breathing now if the rate is regular you'll want to count for 30 seconds and multiply by two however if the rate is irregular you'll want to count for one complete minute so when you go to count the respiratory rate you want to make sure that you were doing it manually rather than going by the number that's on the monitor because as a nurse I have found that that number that that monitor gives you is usually not the rate that your patient is breathing so always double check that now whatever you do go to count the respiratory rate you want to make sure that the patient isn't really aware at that time that you're counting the respiratory rate because they will alter the way that they're breathing it's just like if I told you I'm going to count how many times you blink you're going to alter how much you are blinking you will change it so whenever you do this skill you want to combine it with checking the pulse rate so you'll fill on the side the radial side the pulse rate count that like you normally would but then you're going to keep your finger on that side and count the respiratory rate but the patient thinks that you're counting their pulse rate still and what you're looking for is the rise and the fall of the chest one rise and one fall of the chest equals one breath but this is where the tricky part comes into play because it's like how do I determine what's a rise and what's a fall of the chest so I can count one breath and here are some tips to help you do that now with these tips you're going to find in practice that some of these tips work great on some patients while others you may have to change it up a bit because every patient is built differently they're gonna be wearing different clothes like some patients are wearing downs I may not even have a gown on and some may be wearing their regular street clothes and a lot of patients they have different disease processes going on so one thing I like to do if I know this patient it's gonna be hard to count their respiratory rate I'm gonna get down at the level of their chest like their shoulder and chest area and their abdominal level because I'm looking at that rise and that fall of that chest I'm getting down here looking at that while I'm checking that pulse rate so I start out in that position and some patients it's really easy to tell depending on how they breathe like the shoulder area will move up and down the chest area will move back and forth or the abdominal area will go up and down now if your patient has like a severe disease process where the respiratory system is being affected like with congestive heart failure a lot of fluid volume overload they have trouble breathing so they're going to use these abdominal muscles to help pull in and out that air you can actually just see that I've been going in and out so that's really helpful whenever you're trying to figure out the rate another tip you could do but I would use this with discretion depending on your patient and how they are would be whenever you go to feel that pulse rate you can lightly put your hand on their back on their chest or whatever so you can feel the rise and the fall of the chest but with some patients like if they're agitated confused or they're just someone who he doesn't want you in their personal space you wouldn't want to use this but with some patience you can and my last tip would be to group it with your physical assessment so whenever you're going to assess the chest you're gonna listen to heart sounds and lung sounds what you want to do is tell the patient I'm gonna listen to your heart sounds for one full minute so while you're there listening to both heart sounds and then lung sounds they're not gonna think that something's wrong so you're going to take your stethoscope and first you can listen to their breath sounds and count that rate it's best to try to pick an area that you want to count that rate where there's no a lot of adipose tissue because I can really muffle those breath sounds so generally at that second to fourth intercostal space anteriorly about mid clavicular that could be a great place to count those breath sounds but again it really depends on how your patient is built then after you count that rate then you can proceed and listen to their heart sounds okay so that wraps up this video on some tips for counting the respiratory rate and be sure to check out my other videos such as a head-to-toe assessment how to assess heart and lung sounds along with audio clips of abnormal lung sounds
Nursing_Skills_Videos	Ampule_Medication_Administration_Nursing_Clinical_Skills.txt	hey everyone it's sarah thread sterner sorry and calm and today I'm going to demonstrate how to withdraw medication from an ampule as a nurse you're gonna be required to withdraw medications from glass ampules and you'll want to know how to actually break this ampule without getting hurt and how to remove the liquid medication that is inside of the ampule so you'll want to confirm that you have the right medication and it's the right dose you're giving it at the right time via the right route to the right patient then you'll want to perform hand hygiene and gather your supplies you will need your ampule which contains your medication you also need a syringe that has its own needleless or needle device that you will use to actually administer the medication also you will need a filter straw these come in various sizes 4 inch and 1 and 3/4 we are using a 1 and 3/4 because of the size of our ampule so pick the size accordingly and just like its name says it's a filter straw so we're going to actually attach this device onto our syringe whenever we withdraw the medication out of the glass ampule because it's going to help prevent us from possibly withdrawing in any like glass particles that could go to our patient then we need a gauze because we're gonna use the gauze to actually assist us in breaking this glass ampule and you need some alcohol prep to help you clean the neck of the ampule before you break it first what you want to do is you want to inspect the ampule make sure it's not cracked or has any imperfections also you want to make sure that the solution is clear it doesn't have any particles floating around in it or it's discolored then what you want to do is you want to make sure that all the liquid is down in the body of the ampule because sometimes it likes to collect up here in the head so what you want to do is you just want to hold it steady a little bit and just lightly tap or flick the top of the ampule and the fluid will go down into the body then you want to take your alcohol prep and you want to clean the neck of ampule just to help prevent contamination whenever you go to break that so we clean it and then let it completely dry so while we're letting our ampule dry we're going to connect our syringe to our filter straw and we've already opened our packaging so what we're gonna do is we're going to just unscrew the syringe from its original needleless device and leave it in the packaging because we're actually going to use this device to administer the medication remember you don't administer it with filtered straw and you're gonna take the syringe and just screw it on to the filter straw and leave it in the packaging until we're actually ready to withdraw the medication now we're ready to break open the ampule and many of these glass ampules have a scored area that actually make breaking the ampule a lot easier so if you have that that's great so you're gonna take your clean gauze and you're going to wrap it around the head and the neck of the ampule and whenever we go to break it we're gonna break it this way like in a snapping motion so the broken parts are away from our body so we don't get cut so here we go we're gonna wrap it around the head the neck and we're just going to snap it off then you will have a broken piece of the head and the neck and you want to actually throw this in the sharps container and throw away your gauze now we're going to withdraw the medication out of the ampule so we're going to get our syringe with its filter straw and we're going to insert the straw inside the ampule now you do not have to inject air inside the ampules so make sure your ampule is on a steady flat surface and you're holding it with your fingers to keep it steady I'm gonna go in in the middle and then you're going to which all the prescribed amount of solution so the physician ordered about one cc here so we're gonna pull back until we hit one cc but we're gonna go a smidge over it because we have to remove the air from the syringe and then we'll squirt out the excess of medication then we're going to remove our filter straw and we're going to turn the syringe upright and slightly flick it to get the air bubbles out want to get them all out they can be stubborn sometimes so get it and then you're gonna push up slightly on your plunger and a little bit of medication can sometimes strip out that's okay and you're going to push out to the prescribed amount and then you're going to remove the filters straw and you're going to dispose of the filters straw and the glass ampule into the sharps container after you do that you're going to connect the syringe to its original administration device so just take it screwed on there and then you can leave it in the protective covering until you're ready to administer the medication thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	Cranial_Nerve_5_Examination_Nursing_Trigeminal_Nerve_Cranial_Nerve_V_Assessment.txt	this is hairdresser owner sorry n.com when in this video I want to demonstrate how to check cranial nerve 5 which is the trigeminal nerve so what does the trigeminal nerve do well it's responsible for a couple things number one is responsible for sensation in our face and motor function it allows us to bite down into food so first we're going to test sensation and to do this you're going to need a cotton ball and something with a sharp edge an easy way to get something with a sharp edge is to take a cotton swab cut it in half and this will achieve that sharp edge for you now whenever we're testing for sensation we're testing three areas of the face where the trigeminal nerve branches off and these include ophthalmic maxillary and mandibular regions so first what we're gonna do is we're gonna have the patient close their eyes and we're going to take our cotton ball and we're going to test bilaterally in those areas you just seen and whenever the patient feels that tell them to say yes and what you're looking for is equal feeling what would be abnormal is if they don't feel it the same on each side or they have a decreased response to it so tell me yes when you feel this sensation yes yes yes yes and because he has a beard we're gonna go in this area right here we're not going to go over the hair yes okay now we're going to take our sharp object and we're going to repeat the same way yes yes yes yeah okay and lastly what we're going to do is we're going to test the patient's ability to differentiate between a soft touch and a sharp touch so tell me if you feel it sharp just say sharp or soft okay soft sir soft sure okay now what we're going to do is test the corneal reflex and to do this you'll want to have the patient remove their contacts if they wear them because it can throw off the test you're going to take another cotton ball and you're sort of just going to twist it at the end like this and what we're going to do is have the patient's stare off and you're going to take this part of the cotton ball in at the side of the eye and touch the cornea and what you're looking for is for that eye to blink and simultaneously that other ID to blink as well and if that didn't happen that would be an abnormal response okay so we're going to go in at the side and touch the cornea you should blink now we're going to test motor function and what I want you to do is I want you to bite down for me okay and what you want to do is take your hands and you're gonna feel the masseter muscle in the temporal muscle and you should feel a nice ball of muscle that is equal on both sides and the temporal muscle and I feel that and now normal finding would be if it was unequal on each side or very small next what you want to do is you want to see how well the patient can move their jaw against resistance so I'm gonna have you try to open up your jaw against my hand that was very strong abnormal finding would be they couldn't do that at all and then try to move from side to side okay okay so that is how you test cranial nerve five now be sure to check out my other videos on how you test the other cranial nerves thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	Standing_and_Sitting_with_Crutches_Nursing_Assistive_Devices_NCLEX.txt	hey everyone it's Sarah Reger nurse rn.com and today we're going to demonstrate how to get up and sit down in a chair while using crutches to sit in the chair the patient's going to back up to the chair and fill the chair with the non-injured leg and when the patient fills the chair with a non-injured leg they're going to stop and move both crutches over to the injured side for support then the patient's going to grip the hand grips and slightly bend the non-injured leg and feel behind them and then set in the chair while keeping the injured leg extended to get up from the chair the patient is going to take the crutches and put them on the injured side for support he's going to keep the injured leg extended and push up on the non-injured side and using the hand grips of the crutches then he's going to put the crutches in position thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	MeteredDose_Inhaler_MDI_with_Spacer_Demonstration_Nursing.txt	hey everyone this is cereth registered nurse sorry n.com and in this video I'm gonna demonstrate for you how to use a metered dose inhaler with a spacer okay so what is a metered dose inhaler it is an inhaler that will deliver a set amount of a medication into the lungs so it's used by people who have respiratory diseases like asthma COPD and those medications generally given are like bronchodilators like albuterol or corticosteroids like fluticasone like which is flovent and these medications will help either dilate the lungs which is a bronchodilator or corticosteroid which will help suppress immune system decrease that inflammation in the lungs now when using a metered dose inhaler it is best to use a spacer and why is that well whenever you're using those open or closed mouth techniques like I demonstrated in the previous video those take a lot of dexterity you have to be able to press the inhaler button down and inhale at the same time and if don't do it correctly a lot of the medication will end up in the mouth and instead of the lungs but using the spacer helps decrease that because it gives the patient in a sense time to hit the button and inhale and it helps permit a lot of that medications especially that corticosteroid from collecting in the mouth which can in the end cause thrush so always try to get one of these for your patients so let me show you how to actually use it so the first thing we want to do course is perform hand hygiene perform the patient's five rides make sure you have the right patient the right drug the right time the right route and the right dose because you'll be giving these in pups so check all that information out and another thing you want to remember is while you're doing your check off is that you want to make sure you know exactly which inhaler to give win for instance say you're giving a bronchodilator and a corticosteroid so the patient has two inhalers which one would you want to get first you want to give the bronchodilator first because it's going to dilate out that those lungs and then five minutes later you would want to you're corticosteroid so it can go in there and do its job and decrease inflammation also say that you were just going to be giving one inhaler but it required two puffs well you would give the one dose that they need the one puff and then one minute later you can administer the second puff so if you're giving the same drug you would wait a minute but if you're going to do bronchodilator and then a corticosteroid you would want to wait five minutes in addition after giving the corticosteroid inhaler you would want the patient to immediately gargle and rinse with water and spit that water out to remove any possible corticosteroids that have collected in the mouth because if they stay in the mouth they can irritate those mucous membranes and cause thrush so first of all you want to do is you want to prime the inhaler and you'll do that if this is the first time you're using the inhaler or if it's been dropped or the patient hasn't used it in a week or more if it's been recently cleaned and while you're in the process of doing that just check the inhaler make sure it's not expired you can do that by popping it off the canister and looking at the date and this expires in 2020 so we're good also just quickly check to make sure there's enough doses in this inhaler and of course if it's new there is enough doses but this has been used before you just want to make sure that you have enough puffs in here and a lot of times then hailer has a counter on it that you can look at it or the box will tell you how many doses are in each inhaler so check that out in this particular one there's two hundred sprays so you'll want to count out how many times the patient's going to be using it and how many days that's going to transpire over for instance if the patient was going to be using this inhaler four times a day twice in the morning twice at night that's four puffs and it has 200 puffs how many days is that gonna last us 50 days so you want to keep track of that so to prime it what we're gonna do is we're going to take the cap off and just check inside the inhaler make sure there's nothing hanging out in there sometimes patients like to keep these in their purse and some and other things can get in there and you don't want them to inhale it so make sure it's clean and we're going to give it a good shake so hold it in between your thumb and your fingers and shake it for about eight to ten seconds get it mixed up really well and then we're gonna give it four sprays so it's nice and primed so after priming inhaler you're ready to connect and hailer to the spacer so make sure that the cap of the inhaler is off and then take the cap of the spacer off because this is the mouthpiece where the patient's going to put their mouth at and actually inhale and then this part right here is where you're going to connect the mouthpiece of inhaler into the spacer so just connect it like so and it stays on then have your patient set up hold the inhaler with the thumb and with the fingers and then just give it a good shake for eight to ten seconds then have a patient breathe in through the mouth and then out through the mouth until they can no longer breathe out they will seal their mouth around this chamber with in between their teeth or tongue flat and then they will press the inhaler down and then inhale so it's different from those other techniques that we use where they had to simultaneously press it down and inhale so they press it down then inhale inhale until they can no longer inhale anymore and then they'll hold it for about 10 to 12 seconds and then breathe out slowly and again they would want to rinse their mouth if they use corticosteroids and if they have to repeat the dose they will do that in one minute so now let me show you what that would look like and after that of course you will have the patient rinse their mouth gargle and spit water if it was a corticosteroid and you will recap your inhaler and your spacer then perform hand hygiene and document okay so that wraps up this video on how to use a metered dose inhaler with a spacer thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	PPE_Training_Video_Donning_and_Doffing_PPE_Nursing_Skill.txt	hey everyone it's sarah thread sterner sorry and calm and in this video i want to demonstrate how to dawn and doff personal protective equipment also known as PPE before donning PPE you want to make sure you select the appropriate personal protective equipment and this is based on the potential infectious agent that your patient has plus you want to pick the right size of PPE you want to make sure that you are following the sequence set by your hospital's protocols in this video I'm going to demonstrate just one of the possible sequences you can use to put on PPE so one of the sequences you can use the pin on PPE would be first to put on the gown followed by the mask or respirator depending on what you're going to use then the goggles or face shield and then lastly the gloves and you will Don this PPE before you enter the patient care area first perform hand hygiene then put the gown on by sliding it up the arms and insert the thumbs into the thumb inserts of the gown if present this particular gown has a loop that can be placed over your head then secure the gown by tying the ties in the back next Don a face mask or respirator if using a face mask with ear loops place each loop over the ears then grasp the nose piece of the mask and bring it to cover the bridge of the nose then mold the nose piece of the face mask with the fingertips of both hands by starting at the bridge of the nose and work outward toward the cheekbone then grasp the nose piece of the face mask and pull the bottom of the mask under the chin if using an n95 respirator hold the respirator in one hand with the front of the respirator touching the inside of your hand the metal nose piece should rest near the fingertips be sure the top and bottom straps of the respirator are hanging down below your hand and are not tangled or twisted take the respirator and place it over your nose mouth and under the chin hold the respirator in place with one hand use your free hand to place the straps on your head it's very important the straps rest on your head and are not overlapping or first take the top strap of the respirator and place it just past the top of your head above the ears then take the bottom strap of the respirator and place it around the bottom of the head just below the ears and then make sure the straps are not overlapping or twisted to ensure a proper seal of the respirator around the nose area use the fingertips of both hands to bend the metal nose piece around the nose starting in the center and working your way outward on each side now perform a seal test place both hands over the respirator take a sharp breath in and make sure that the respirator is sealing over the face during this inhalation keep both hands on the respirator and breathe out while breathing out feel for any air leaking around the nose piece and around the respirators perimeter if air is leaking around the nose piece try rebuilding the nose piece with both hands and recheck the seal if air is leaking around the respirators perimeter try adjusting the straps and recheck again making sure the respirator is not being affected by any piercings facial hair or glasses which could prevent a proper seal if the respirator still leaks after this you may need a different size or type of respirator do not use the respirator if a proper seal has not been obtained next dawn a face shield or goggles if using goggles apply goggles over the eyes make sure any straps or bands of the goggles are secured around the head and the goggles fit snugly but comfortably then perform hand hygiene again lastly dawn gloves by pulling each glove over the hand and extending the cuff of the gloves over the sleeves of the gown now I want to show you when the possible sequences you can use to Dolf PPE one of the sequences you can use to remove PPE would be first removing the gloves followed by the gown then the face shield or goggles then lastly the mask or respirator when removing the PPE you will be doing this inside the patient care area however if you wore a respirator the respirator will be removed once you leave the patient care area first remove the gloves do not touch the outside of the gloves because they are considered contaminated to remove the gloves start by taking your non-dominant gloved hand to grab the other glove around the cuff in the wrist area do this by using a pinching motion to grab it peel this glove off by turning it inside out and what it into a ball with your gloved hand keep it securely in your gloved hand then take the index finger of the unloved hand and slide it carefully under the cuff of the gloved hand and peel the glove off the hand by pushing the index finger forward against the glove this will turn the glove inside out dispose of the gloves next remove the gown the front of the gown and sleeves are considered contaminated whereas the ties on the inside of the gown are considered clean therefore avoid touching the front of the gown and sleeves carefully untie or break the ties of the gown peel off the gown by removing the thumbs from the inserts if present and then move the shoulders and arms up through the gown which helps allow the gown to slide off the body then from the inside with the assistance of the arms roll the gown away from the body by turning it inside out form it into a ball and discard then perform hand hygiene next remove the face shield or goggles do not touch the outside of the goggles or face shield because they are considered contaminated if goggles are used remove the goggles by carefully grasping and lifting the back band of the goggles over the head dispose or clean the goggles / facilities protocol then remove the mask or respirator if a mask is worn the front of the mask is considered contaminated therefore grasp the ear loops of the mask with the fingertips and remove the mask from the face discarded the mask / facilities protocol and perform hand hygiene if a respirator is worn the doffing procedure is performed outside the patient care area the front of the respirator is considered contaminated therefore removed by the straps grasp the bottom strap of the mask and lift it over the head then grasp the top strap of the mask and lift it over the head dispose or reuse the mask / facilities protocol then perform hand hygiene okay so that wraps up this video on how to don and doff PPE and don't forget to check out the other videos in this nursing skills series
Nursing_Skills_Videos	Cranial_Nerve_Examination_Nursing_Cranial_Nerve_Assessment_IXII_112.txt	this is Sarah with register nurse rn.com and in this video I'm going to be going over a nursing assessment of the cranial nerves cranial nerves 1 through 12 and if you would like to watch a complete head-to-toe nursing assessment you can access this card up here in the corner or in the YouTube description below to watch that video so the first thing what you want to do is you want to provide privacy to the patient perform hand hygiene and explain what you will be doing so let's get started we're going to test the olive Factory cranium nerve one the sense of smell so Ben what I'm going to have you do is I'm going to have you close your eyes and I'm going to put something in front of your nose and have you breathe in and smell and you tell me what you smell and whenever you do this use something that's Pleasant smelling not something that's really stinky because it could elicit like a gag reflex or something like that if the person has a sensitive nose Okay vanilla okay and this was vanilla extract and that is correct so that cranial nerve is is intact to test cranial nerve 2 you're going to be doing two tests number one you're going to look at the confrontation visual field and you're going to be looking at their peripheral vision next you'll be looking at visual Acuity and using a smelling chart to assess that first we're going to test the peripheral vision by doing the confrontation visual fill test and to do that you're going to have the patient stand in front of you about arms length away okay Ben what I'm going to have you do is I'm going to have you cover up your right eye and I'm also going to cover up my left eye so on the same side and B I want you to look at this eye and don't look at my fingers okay and I just want you to tell me how many fingers you see and you're going to do this in the upper and lower visual fields in about the middle of the visual field so here we go two correct three very good okay let's do it on the opposite again stare at this eye okay and don't look at my fingers three one very good now we're going to test visual Acy using a Snelling chart and what you're going to do is you're going to have your patient stand about 20 feet from the chart so Ben if you'll stand about right there for me and ask your patient do you wear glasses no okay and if your patient does wear glasses you'll want them to wear those for this test okay so what we're going to do look at that chart over there and try to read the lowest line for me that you can read okay okay and first we're going to cover your right eye then your left eye and then we'll do both eyes Okay so cover your right eye okay and what line can you read eight okay read it for me d e f p o t e c okay very good okay now we're going to cover up your left eye and do the same thing and again whatever line you can read let me know eight eight again okay d e f p o t e c okay and now read with both eyes and what line a okay d e f p o t e c okay and he read from line eight so that means that he has 2020 vision and this means that he can see the same line of letters at 20 feet that a person with normal vision can see at 20 feet however let's say that in his left eye he could only read like line six which is 2030 that would mean that his left eye sees at 20 ft that a person with normal vision would see at 30 next what we're going to do is we're going to assess cranial nerve three which is ocular motor four tular and then six which is abducent and we're going to do several tests to check their function the first one what we're going to do is we're going to be looking for any involuntary shaking of the eye called nagas and how we're going to do that is we're going to take our pin light we're going to hold it about 12 to 14 inches away from the patient's nose and Ben what I want you to do is keep your head still don't move your head and just use your eyes to watch where I move the pin light and as you're doing this you're going to do you're going to perform it in the six cardinal fields of gaze and and you're just going to move it and you're looking for any involuntary shaking of the eyes so here we go next we're going to see how reactive the pupils are to lie and to do that we're going to dim the lights a little bit and we're going to have the patient stare off at a distant object that helps dilate those pupils and then we're going to shine using our pin light in at the side and we're going to see how that pupil responds it should constrict and then on the other side it should constrict as well so say their Baseline pupil size was like 3 millim it should go down to 1 ml and it should happen on both sides okay so B stare off at that object right on the wall over there for me okay and that dilates the peoples and we're just going to shine light in at the side okay okay constrict constrict okay let dilate again then go to the other side do the same again and they both constricted in equal size next what we're going to do is we're going to check for accommodation and how we do that is we turn the lights back on we just previously had them dim but we now make it light again we're going to have him stare off at a distant object that helps dilate the pupils and we're going to take a pin light you can use a pin light finger and you're just going to slowly move it inward to the nose and what you're looking for is that those pupil's constrict they accommodate and the eyes cross while looking at the pin light so here we go stare off in the distance please and I don't want you to move your head or anything just keep it real still and just follow this pin light okay ready okay so now we can document cuz we just checked all of the things with the eyes we can document that the pupils are equal round reactive to lie and accomodate so that's where that acronym p e r r l a comes into play we're going to go ahead and test cranial nerve five which is the trigeminal nerve and this nerve is responsible for many things like mastication so what I'm going to have you do B is I'm going to have you clinch your teeth like bite down for me and I'm going to feel the Meer muscle which is right there and it should be a nice firm ball and then feel the temporal muscle now what I'm going to do to also test that nerve is have him try to open his mouth against resistance so try to do that for me okay and he can do that and while we're here we're going to go ahead and look at the facial expressions and test cranial nerve 7even which is the facial nerve so can you close your eyes tightly for me and open them up okay now smile for me friend and puff out your cheeks okay and he did that with e so that cranial nerve is intact next we're going to test cranial nerve eight which is the vestibulo cular nerve and what I'm going to do is I'm going to include one of his ears and then whisper two words on the other side he needs to tell me what I said so you ready mm okay I'm going to include this one apple banana apple banana okay very good cat dog cat dog okay and that nerve is intact next what we're going to do is we're going to assess cranial nerve nine the glosso faren gel so what I'm going to do is I'm going to have you say ah and what you want is that uela to move up ah okay and then we're just going to test the gag reflex I'm sort of just going to poke a little bit back there and elicit a gag CL okay there you go gags really good and um cranial nerve 10 the Vegas is intact because he's able to talk with talk to me without heness and he's able to swallow then what we're going to do is we're going to test cranial nerve 11 which is the accessory nerve so being what I'm going to have you do is move your head side to side up and down okay and then shrug try to shrug against my resistance and he does that with ease so that nerve is intact next what we're going to do is we're going to a cranial nerve 12 which is the hypo glossal nerve and what I'm going to have you do Ben is I'm going to have you stick out your tongue and move it side to side okay and he does that with ease okay so that wraps up how to perform a nursing assessment of the cranial nerves and don't forget to check out that head to toe nursing assessment video thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	4Point_Gait_Crutches_Walking_Pattern_Demonstration_Nursing_Skill.txt	hey everyone it's Sarah threads sterner sorry and calm and today we're going to demonstrate how to do the four point gate while using crutches so the patient will move the right crutch witches will say the injured side then they'll move the left foot then they'll move the left crutch and then they'll move the right foot thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	Temporal_Pulse_Point_Palpation_Location_and_Nursing_Assessment.txt	hey everyone it's sarah thread sterner sorry and calm and today I'm going to demonstrate how to find the temporal pulse point and whenever you're assessing the pulse you will be looking at a few things one thing will be the rate how fast is it along with the strength and you'll be grading it on a scale zero to three with zero being absent one plus it's week two plus it's normal and three plus its bounding and then you'll want to look at the rhythm is it regular or irregular and as you fill on the pulse you will be filling bilaterally to see if they're equal and to find this artery you'll want to find landmarks and this always goes for any pulse point that you're trying to find now this temporal artery comes off of the external carotid and goes up and what you want to do is you want to find the tragus of the ear which is this part of the ear and the zygomatic arch is found above it which is a fancy way of saying the cheekbone so your pulse site where this X is is found right here this is your temporal artery and you'll want to feel bilaterally and see if they are equal and the time that a nurse really feels on this is during that head-to-toe assessment when whenever we're assessing the pulses in the head okay so that wraps up this video on how to check the temporal pulse point and don't forget to check out our other videos on how to assess other pulse points on the body thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	How_to_Mix_Insulin_NPH_and_Regular_Insulin_Nursing_Mixing_Insulin_Clear_to_Cloudy.txt	hey everyone it's Sarah register nurse rn.com and in this video I'm going to be going over how to mix insulin what I'm going to do for you is that I'm going to actually show you how to draw up insulin whenever you're mixing it but first let me go over the most important Concepts that you need to know whenever performing this nursing skill so what is the purpose of mixing insulin well the purpose is is that it helps prevent from having to give the patient two injections because a lot of times Physicians will order two different types of insulins and you can just mix them in the syringe and give them one injection so it is better for the patient now in this video what I'm want to be going over are the most commonly ordered insulins that you mix which generally are NPH which is an intermediate insulin and regular insulin which is a short acting insulin now before we get started let's go over these key Concepts that you need to remember number One never mix lantis which is a long acting insulin with any other type of insulin it's to go all by itself so you will never mix it with anything number two after drawing up the insulin into your syringe administer the dose within 5 to 15 minutes because regular insulin binds to mph and it will decrease its action number three before administering insulin always check the patient for signs and symptoms of hypoglycemia so you want to look at the patient you want to ask them how they're feeling because typical signs and symptoms of hypoglycemia is that the patient's sweaty clammy Tac cardic confused and a lot of times patients will know they'll say my blood sugar feels low can you check it because they know their body a lot better than we do and the reason you want to check their blood sugar ask them assess them for signs of hypoglycemia before you give insulin because the way the insulin works is that it will take the sugar that's in the body and transport it into the cell so it deplete the system of glucose so if you're giving them more when it's already depleted you're going to cause some really bad problems and then check their GL their glucose depending on your hospital guidelines whatever the glucometer readings um whatever is considered hypo and hypoglycemia check for that but generally anything less than 70 and if that happens what you would want to do is that you would want to hold that dose and notify the doctor immediately and ask what the further orders would be another important concept if you don't get anything out this whole video with um how to mix insulin this is important okay how you draw up the insulin you're going to have a cloudy solution which is your mph the intermediate and you're going to have a clear solution which is the regular acting insulin so whenever you look at your vs that's what it's going to look like now you're going to draw up in this order clear to cloudy the clear again is regular insulin and the Cloudy is n pH the intermediate active insulin now to help you remember this because a lot of people get this confused try to remember the pneumonic R in registered nerves R for regular and N for MPH so that's how you're going to draw up why do you do this what's the purpose of drawing it up like this well it prevents contaminating the Cloudy insulin which is the intermediate insulin with clear insulin which is the regular insulin now if that happen it would decree it would affect the action of how that mph insulin would work so you don't want to contaminate the Cloudy with the clear so that's why we draw it up that way now I'm going to show you how to actually mix insulin now this is just an example always follow your hospital protocols your doctor's orders because procedures and medications do change over time okay the very first thing you want to do is you want to check your Physician's order to confirm how much insulin you're actually giving and our doctors order says administer 10 units of humin R and then 12 units of humin in subq before breakfast daily after you confirm that you want to make sure you have the right medications so you're going to take your vial and you're going to look and read what's on the vial here we have humin R and you can tell because it's a clear solution and then we have our other vial which is humin in and you can tell because it is cloudy and you need to ask yourself how many ins how many total units of insulin am I going to be giving um 10 units of humin R and 12 units of humin n so that equals 22 units total then you want to make sure you know the peak times be familiar with the peak times of each of these insulins I have a video of newonics that can help you remember the peak times of this a card should be popping up and you can access that so let's go over it real fast the peak of humin R is 2 hours and the peak of humin n is 8 hours and this is when the patient is at most risk for hypoglycemia inlex Li likes to ask about that so make sure you know that then next what we're going to do is we are going to perform hand hygiene clean our hands and we're going to put on some gloves then after we put on our gloves we are going to take our cloudy insulin because this is a cloudy solution it's their mph the intermediate insulin and we need to mix these ingredients cuz they like to settle so to do that what you're going to do is you're just going to roll the insulin gently in between the palms of your hands and just roll it and get it mixed up because if you don't do this it'll alter how much mph you're actually drawing up so you want to make sure the solution is good mix and never shake it because that causes air bubbles and that can throw off how much mph you will actually draw up okay next what we're going to do is we're going to take our alcohol prep and we are going to clean the tops of these vials make sure we're getting rid of germs so we're just going to clean for about 5 to 10 seconds really good each file and then we're going to go to the next file and just clean that and now we are ready to use our syringe so what we're going to do is you're going to take the cap off of your syringe and we are going remember remember the pneumonic RN we're going to start with regular insulin and then go to cloudy so it's going to go clear to cloudy regular to mph so we are going to inject 12 units of air and our humin in insulin so just pull back on your syringe to 12 units which is there we're going to inject the 12 units of air into the vial then we're going to remove our syringe and we're going to inject 10 units of air into the humant R so we're going to pull back on the plunger to 10 units and we're going to do the same thing but we're not going to remove our syringe instead we're going to take the bottle and flip it upside down and we are going to remove 10 units of regular insulin and you watch on your syringe where 10 units is and make sure it's precise and once you have that 10 units you're going to remove the vial and then we're going to go to our humin because that was the clear now we're going to cloudy and we're going to remove a total of 22 units for the whole total dose so going to go into our M and we're going to remove a total of 20 2 units that will equal how much we're to give now after you draw it up if you're not going to give it immediately what you want to do to prevent contamination sticking yourself or sticking someone else is to do the onand scoop technique so just take it and take the cap and just scoop it up and then you are ready to go okay so that is how you mix insulin now be sure to check out my inlex review on diabetes and check out my other nursing skill videos and thank you so much for watching and please consider subscribing to this YouTube channel
Nursing_Skills_Videos	How_to_Stand_and_Sit_with_a_Cane_Nursing_Assistive_Devices_Skill_Review.txt	hey everyone its era threads owner sorry ENCOM and in this video I'm going to demonstrate how to get up and sit down in a chair with a cane to sit down with a cane the patient is going to back up to the chair until he fills the chair with the back of the legs then the patient will allow the cane to rest on the side of the chair and place both hands on the chairs armrest and place weight on the hands while keeping the weak leg extended out and been the strong leg to sit down to sit up with the cane a patient's going to place the cane on the strong side and lean forward in the chair while keeping the weak leg slightly extended forward then the patient's going to push down on the canes hand grip and the chair armrest and then put weight on the strong leg and stand in position with the cane thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	Nitroglycerin_Medication_Nursing_Sublingual_Tablets_Oral_Spray_Pharmacology_Review_Adminstration.txt	this is Sara threats Turner sorry and calm and in this video I'm going to be going over sublingual nitroglycerin and the oral spray and I'm going to show you how to administer it now remember whenever you administer nitroglycerin you always want to follow your hospital protocols for this medication along with the manufacturer's instructions and always be sure you have the proper credentials when administering this medication and for this video we are using demo dose doses of nitroglycerin so it's a simulated dose and these do not really contain nitroglycerin so what does nitroglycerin use for it's used to relieve or treat chest pain and nitroglycerin is part of the nitrate family and because of this what it does is it causes vasodilation on those blood vessels so you have a patient they're having chest pain they're getting limited amount of blood to that heart muscle and what nitroglycerin will do open up those vessels and increase the blood supply to that heart tissue along with decreasing its workload so as a nurse what do you expect to see in your patient after you give them this drug well you can expect to see facial flushing from where they're getting increased blood flow through the body from that vasodilation they can also experience like a burning or tingling sensation underneath the tongue whenever you give this sublingual the tablets you will put underneath the tongue and the spray you will spray under the tongue as well also it can cause a very painful headache from all the extra blood flow they can experience dizziness and definitely hypotension it's going to lower the blood pressure so when your patient is having chest pain as the nurse of course you want to notify the physician and a lot of hospitals already have standing orders there for you to follow out they're usually called chest pain protocols and you will have this order set that you need to do for your patient which will include you know giving the nitroglycerin possibly morphine if the Nitro is not really helping along with oxygen administer via nasal canula of 20 obtaining a 12-lead EKG and drawing cardiac enzymes so before you administer a nitroglycerin to your patient of course you want to confirm that they are not allergic to it and that they haven't recently taken a phosphodiesterase inhibitor like sadena fill which is viagra or $2.00 fill which is cialis because this can lead to severe hypotension and possibly even death and of course you don't want to give this to a patient who has increased intracranial pressure then you'll want to perform hand hygiene perform the patient's 5 right and whenever handling nitroglycerin you want to wear gloves because if you get this on yourself you can give yourself some nitroglycerin and you can get some of the side effects that come along with this drug so first let's talk about sublingual nitroglycerin okay it comes in a dark vial and this vial is dark for a reason because nitroglycerin is very sensitive and it doesn't need to come into contact with white or heat or moisture and it's only good for three months when the bottle is opened and then you're gonna have the patient sat down before administering this to them because remember it can cause dizziness and hypotension and you want to obtain baseline vital signs especially blood pressure because you're gonna see a decrease in their blood pressure and you want to confirm that their systolic blood pressure is within parameters and most Hospital protocols require that that's a solid blood pressure that top number is greater than 90 and if it's not and your patients having chest pain you need to contact the physician for further orders and preferably you want them on a cardiac monitor so you can be monitoring that rhythm looking at the ST segment for normalities like St elevation or depression to administer you will place one tablet under the tongue so how the patient lifts the tongue up and it will dissolve under there the patient does not need to chew it or swallow it and not to rinse the mouth because they may feel like they need to do that whenever they have the tablet under the time because of the tingling sensation and it's going to absorb within that lining there and after giving the tablet you'll want to make sure that you're constantly monitoring their blood pressure especially that's systolic making sure it's greater then 90 and their chest pain how is it rated is it increasing is it decreasing is it going away and if their systolic blood pressure is still within parameters and they're still having chest pain you can do a second dose within five minutes again monitoring that blood pressure and their chest pain and the chest pain is still there you can give a third dose within five minutes now you do not want to get more than three doses and if the chest pain is still not relieved after the third dose you'll want to notify the physician now with the oral nitroglycerin spray what you want to do is of course remove the cap on the bottle and if this is a new bottle you'll want to prime it and a lot of times in the hospital setting and you're going to be using a new bottle of nitroglycerin and to prime it you're going to press the button to spray about five to eight times just to get that line in there nice and primed and whenever you do that tip keep it away from your face and other people's faces you don't want to inhale this so we're gonna probably about five to eight times all right okay and then it's ready to use and another tip with this spray is that you don't want to shake the bottle up and down remember nitroglycerin is very sensitive and it's in a dark bottle so keep it away from heat and light and keep it in the upright position at all times you don't want to store it upside-down and to give the world nitroglycerin spray you're gonna give one spray underneath the tongue and tell the patient to slightly hold their breath and not to breathe in the medication now after giving the oral spray you'll want to make sure that the patient doesn't rinse their mouth or eat or drink anything for at least ten minutes and you'll follow the same protocol as you followed with the sublingual tablet by monitoring that blood pressure constantly making sure it's within parameters and monitoring their chest pain and within five minutes you can give a second dose and within five minutes you can give a third dose and no more than three doses okay so that is how you give sublingual nitroglycerin and oral spray thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	Chest_Assessment_Nursing_Heart_Lung_Assessment_HeadtoToe_Exam.txt	this is cereth registered nurse orient home and in this video i'm going to demonstrate how to assess the chest and if you would like to watch a complete nursing head-to-toe assessment you can access this card up here in the corner or in the youtube description below now to do this skill what you'll need to do is you'll need to perform hand hygiene provide privacy to the patient and tell them what you will be doing and you'll need to grab your stethoscope so let's get started we're going to inspect the chest we're looking for any abnormalities like lesions or any wounds anything like that we're also inspecting the patient's effort of breathing is it really labored are they using those accessory muscles to breathe also we're looking at that anterior posterior diameter so turn to the side like that and you're looking for that barrel chest and can will be increased in patients with like COPD they will have what's called the barrel chest and now what we're going to do is we're going to listen to heart sounds and then we're going to listen to lung sounds so first let's auscultate heart sounds and we're going to do this in five locations and they're based on where the valves are located and I like to remember the mnemonic all patients effectively take medicine and the first letter of each word represents the valve except for effectively so a would be a or tick p and patients would be pulmonic effectively would be herbs point and this is just the halfway point between the base of the heart and the apex of the heart and there's no valve location there and then t is for tricuspid and then n is for medicine so again using the diaphragm we're going to listen at the right of the sternal border at the second intercostal space and that's going to be the aortic valve so to find that second intercostal space find the sternal notch go down to the angle of Luis and then just go a little bit to the right and you're in the second intercostal space and this will be the aortic and we're just listening lub-dub lub-dub s1 s2 and s2 the dub is going to be louder in this location then we're going to go a little bit over to where the pulmonic valve is found that's on the left of the sternal border at the second intercostal space so we were just right across Kandace listening to love dub lub dub and s2 dub is going to be louder in this location then we're going to go a little bit down to the third intercostal space and this is herb's point and again and here love dub that there's no specific valve here then we're going to go down to the fourth intercostal space and this is where the tricuspid valve is and love s1 is going to be the loudest at this location then we're going to go to the fifth intercostal space midclavicular line and we're going to listen to the mitral valve and again s1 is going to be loud us hear dub and there's something special about this site this is the point of maximal impulse this is where you're going to listen for the apical pulse so we're gonna set here and we're gonna counter it for one full minute and a normal apical pulse an adult should be 60 to 100 beats per minute and his apical pulse was 63 then we're going to switch over with the Bell of our stethoscope and we're just going to repeat in those locations and we're specifically listening for heart murmur so that's swishing blowing sound so that's what we're going to listen to with that and I did not hear any now let's listen to lung sounds now when you're listening to lung sounds you're listening for abnormal sounds and here are some samples of some abnormal sounds that you may hear crackles wheezes a friction rub or Strider first we're going to listen and tearily and what we're gonna do is we're gonna listen with the diaphragm over stethoscope and we're gonna start at the apex of the lungs and we're gonna always compare sides and just enter way downward and assess all the lobes of that right and left lung so first let's start up here okay and we don't want you take good deep breath in and out so here we go apex okay we're gonna come here sides then we're going to move down to the second intercostal space and this is going to help us assess the right upper lobe and the left upper lobe so another deep breath then we're going to go down to the fourth intercostal space and we're going to assess where our right middle lobe is and our left upper lobe because remember the right lung has three lobes and the left lung has two lobes so let's listen to our left upper lobe a little bit more then we're gonna go mid-axillary at the six six intercostal space and we're gonna listen to the right and left lower lobe so you just want to turn to the side right there okay other side okay now let's listen post eerily again using the diaphragm here the stethoscope you're going to start listening at the apex and work your way down and one thing to keep in mind when you're listening back here you have the scapula and you don't want to listen over those because you won't be able to hear the sound so you're gonna listen in between where the scapula and the spine are so down in these regions right here again we're just going to compare sides and you can do this part at the end if you wanted to whenever you turn your patient over to look at their back side but we're just going to go ahead and do it now so we're gonna start and the apex compared sides then we're going to find c7 which is that vertebral prominence it's the big ball right there you can't miss it and go down to about t3 and you'll be in between the shoulder blades and go a little bit in between the shoulder blades and the spine right in there and you're gonna assess the right and left upper lobes then from t3 to t10 we're just going to inch around and we're going to listen to the right and left lower lobes okay so that is how you assess the chest and don't forget to watch that video on a complete head-to-toe nursing assessment thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	Lung_Sounds_Normal_Bronchial_Bronchovesicular_Vesicular_Breath_Respiratory_Sounds.txt	hey everyone this is Sarah register nurse rn.com and in this video I'm going to let you listen to Three normal lung sounds this will be part of a review series of the lungs I'll be doing a video on abnormal lung sounds as well as a lecture and skills demonstration video as part of this series so let's get started and okay thank you so much for watching remember you can go to our channel to find more videos on nursing skills inlex prep and more so please subscribe and share this video with others
Nursing_Skills_Videos	Dorsalis_Pedis_and_Posterior_Tibial_Pulse_Point_Nursing_Assessment.txt	hey everyone it's sarah thread sterner sorry and calm and today I want to demonstrate how to find the dorsalis pedis and posterior tibial pulse points and whenever you're assessing the pulse you will be looking at a few things one thing will be the rate how fast is it along with the strength and you'll be grading it on a scale zero to three with zero being absent one plus it's week two plus it's normal and three plus its bounding and then you'll want to look at the rhythm is it regular or irregular and as you fill on the pulse you will be feeling bilaterally to see if they're equal now we're going to find the posterior tibial and the dorsalis pedis and these sites are assessed during your head-to-toe assessment especially if your patient might have peripheral vascular disease or if they've had a vascular procedure like a heart cath where they access the femoral artery we'll always be checking the pulses in these lower extremities now the dorsalis pedis want to turn the fit a little bit this way is found here and in order to find that what you can do is find the e HL tendon the extensor helices longus tendon that actually helps extend this big toe so if you've lift your toe it for me you can see that tendon dry in here now just follow that tendon and then just go a little bit to the end of it and that will be your dorsalis pedis then we'll find the posterior tibial and this is found on the inside of the ankle found between the back of the medial malleolus which is the bony prominence of the ankle bone and Achilles tendon so you'll find it within this area right here okay so that wraps up this video on how to check the pulse points of the fee and don't forget to check out our other videos on how to check the other pulse points of the body thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	How_to_Sit_Down_and_Stand_Up_with_a_Walker_Sit_to_Stand_with_Walker.txt	hey everyone it's Sarah threads to nurse our end comm and in this video I want to demonstrate how to get up and sit down in a chair with a walker to sit down in a chair a patient is going to take their Walker hold on to the hand grips and slowly back up to the chair until they fill the chair with the back of their legs then they're going to slightly extend that weak leg out and take their hands and position them behind them and bend their strong leg and feel for the chairs armrests and then set down to get up from the chair what the patient's going to do is of course make sure the Walker is out in front of them they're going to lean forward out of the chair make sure their hands are on the hand rest and slightly extend that weak leg out then they're going to put weight on their hands by pushing up on the armrest of the chair and with their strong leg and putting their hands on the hand grip of the Walker then they are ready to ambulate and again to do that they will lift the Walker make sure all points are on the floor move that weak leg put weight on the handgrips and then move the strong side thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	Incentive_Spirometry_Spirometer_Demonstration_Instruction_Incentive_Spirometer_Procedure.txt	okay first let's go over some wrong ways to use in some spirometer a lot of times if patients haven't been properly they will do what they think you're supposed to do naturally with it by blowing into the device rather than actually inhaling from the device another wrong way patients may try to use in sims from--it ER is by quickly inhaling and exhaling off the device like this now let's look at the right way to use incentive spirometer okay first what you want to do is you want to set the goal for the patient with the yellow marker so they know where they need to get whenever using this instead of spirometer then you're going to have the patient set up and exhale completely then have them seal their mouth around the mouthpiece tightly and they will inhale slowly and deeply making sure to keep the yellow indicator on the side within normal range they don't want that little yellow piece to go too high or too low and as they do this the piston will rise up and have the patient keep inhaling as deep as possible until they can't inhale anymore and then they'll need to hold the breath for six seconds and then exhale slowly and allow the piston to fall before repeating again and record the amount that they were able to get on incentive spirometer and they will perform this at least ten times every hour to wile away so this is what it looks like in action
Nursing_Skills_Videos	Newborn_Reflexes_Assessment_Infant_Nursing_Pediatric_NCLEX_Review.txt	hey everyone it's sarah with registerednessrn.com and in this video i'm going to demonstrate how to assess newborn reflexes so let's get started newborn reflexes are reflexes that a newborn baby is born with and the reason we care about newborn reflexes is because it tells us how well that nervous system is developing and functioning these infant reflexes also called primitive reflexes will actually disappear over a certain time period which is a good thing because it tells us again that this neurosystem is developing like it should therefore for exams you want to know three things first you want to know the names of the main reflexes that you're going to assess in a newborn second you want to know how to get a response out of those reflexes and what an appropriate response should be and then third you want to know about the approximate time when these reflexes should disappear so now let's look at these reflexes the first reflexes we're going to look at are the grasp reflexes and this includes the palmer and the plantar grass reflex so first the palmer grasp palmer means we're talking about the hand so this reflex deals with something with the hand so you can get a response out of the infant whenever you place a finger or you stroke the inside of the infant's palm and whenever you do this the hand will close around it this tends to disappear around about four to six months of age then we have the plantar grasp and plantar deals with the foot so you get a response from the infant with this whenever you place a finger underneath the toes and whenever you do this the toes will actually curl like they're grasping the finger this disappears around about nine months to one year of age next is the moro reflex and this is also sometimes called the startle reflex so this is seen when an infant hears a sudden loud noise they will do this reflex or if they experience an unexpected movement like if you go to lay them down and you lay them down where they're supine and you're supporting their head and their heads just a little bit lower than their body and you lay them down on a surface they will do this reflex so what it looks like is that the infant will throw out their arms hence extend them and the palms will be upward sometimes they'll move the arms back to the body after doing this and they may even cry and this reflex tends to disappear around about six months of age then we have the rooting reflex and this is seen when the infant's cheek or side of mouth is stroke the head will actually turn towards it and the infant's mouth will open in an attempt to suck so this reflex is really helpful in helping the baby find their food source when they want to feed and it tends to disappear around about four months of age next is the sucking reflex and just like the rooting reflex the sucking reflex helps the baby feed it's going to play a role with nutrition so you will see this whenever something touches the top of the infant's mouth on the inside it's specifically that hard palate of the mouth so the roof of the mouth the infant will actually automatically just start sucking it's an involuntary thing that happens so again this helps the infant with feeding by allowing the infant to pull food from its food source like the breast or a bottle now this reflex tends to disappear at about four months and it will actually become a voluntary function then there's the babinski reflex and this is seen when the bottom of the foot is stroked from the heel upward along the outward part of the foot so what's going to happen with this you will see the big toe dorsiflex so it's going to bend backwards and the other toes are going to fan out or spread out and this is what you want to see in an infant now in an adult you do not want to see this response i actually have a video where i show you what a normal and abnormal babinski reflex is in an adult and you can check that video out if you're interested and this tends to disappear this type of response around about one year of age the next is the crawling reflex also sometimes called bower crawling reflex and how you get a response out of an infant with this is that you place them on their stomach and you apply pressure with your hand to the sole of the foot so whenever you do this the infant will attempt to push against thy hand with their foot and move the arms and legs in a crawling like motion and this tends to disappear anywhere between a few weeks after birth to a couple months next is the step reflex and just as the name of this reflex says this deals with the infant taking steps so this is seen whenever you hold the infant upright with the legs and the feet touching a surface the infant will actually move the legs like they're taking steps or walking and this tends to disappear anywhere between three to four months of age and lastly we have the tonic neck reflex and this is sometimes referred to as the fencing reflex and let that name fencing reflex help you because it'll help you understand how this infant will look whenever this reflex is initiated because fencing is a type of sport where people fight with swords so the response is whenever an infant's head is turned to a particular side the leg and arm on that side will extend while the leg and arm on the opposite side will flex similar to how those people look whenever they're fencing and this reflex tends to disappear at about four months of age okay so that wraps up this video over newborn reflexes and be sure to check out the other videos in this pediatrics series
Nursing_Skills_Videos	HeadtoToe_Assessment_Nursing_Nursing_Physical_Health_Assessment_Exam_Skills.txt	this is sarah with registerednessrn.com and in this video i'm going to be performing a nursing head-to-toe assessment this video will be similar to what you have performed in nursing school whenever you're doing your clinical check-off now whenever you actually start working as a nurse you'll be able to tailor this head to toe assessment to focus on the patient's needs and you'll get a lot faster at this so what i want to do is i want to cover literally how to assess from the hair on the head all the way down to the toes so let's get started now when you're doing your head to toe assessment you follow that sequence of how you assess each system so you start out whenever you're looking at a system you're going to inspect it then palpate percuss and then auscultate except you're going to change it up a little bit whenever you're going over the abdomen you're going to inspect auscultate percuss and then palpate and the reason that you're going to auscultate second instead of last is because whenever you perform palpation percussion if you did that before it could alter the bowel sound so we want to go ahead and just auscultate get a baseline of what we can hear and then we will percuss and palpate so first what you want to do is you want to perform hand hygiene and provide privacy to the patient then introduce yourself to the patient and explain what you're going to be doing so hello my name is sarah and i'm going to be your nurse today and i need to perform a head to toe assessment is that okay with you okay then proceed and look at their armbands so while you're doing this this is going to help you make sure you have the right patient and you're going to be testing them to see if they know who they are their date of birth and ask them some other questions to assess that neurostatus so say your first and last name for me first name is ben last name is dover okay and your date of birth 8 28 82. okay and do you want me to call you ben or mr dover what do you mean okay so ben can you tell me where you're at i'm at the hospital okay and can you tell me what we're doing here today uh head to toe assessment okay and who's the president of the united states donald trump okay so he answered all those correctly and he's alert and he's oriented times four he knew who he was he was able to tell me his name his date of birth where he was what we're doing and current events so we can chart alert and oriented times four then you want to collect vital signs such as the patient's heart rate blood pressure temperature oxygen saturation respiratory rate and the patient's pain rating so ben are you having any pain on a scale zero to ten with zero being no pain at all and ten being the worst pain you've ever had no pain zero okay and i have a video on how to assess those vital signs in depth if you want to watch that video and a card should be popping up so you can access that video then after that what you want to do you can collect their height and their weight and look at the bmi their body mass index remember if it's 18.5 or less that's underweight or if it's greater than 30 that is obese now while you've been doing all that asking them these questions doing their vital signs you're also before you've even really the system you are already collecting information for instance how is that patient responding to you what's their emotional status are they calm are they agitated are they drowsy what's going on with them do they look their stated age does his skin color match his ethnicity is he does he understand my questions or does he seem like he can't hear them very well or is there a delay whenever he responds to me or does he respond appropriately and at an appropriate time also do you notice any just outward abnormalities like an amputation any masses lesions is his skin sweaty cold and clammy do you notice any cyanosis right off the bat also do is his hygiene good and is his posture good and do you notice any abnormal smell so during all that you're really collecting all that information now what we're going to do is we're going to start with the head and move our way down to the toes so we are first going to inspect the head and we are looking at the skin color he it's nice and pink we're also going to make sure that the head is the same size as how it should be for the body and it is and we're looking for any abnormal movements or twitching of the face that he can't control that are involuntary we don't see anything and we're making sure that the face is symmetrical there's no drooping on one side like in this picture there's drooping on one side of the face and this can be seen in bell's palsy or in stroke and we're also just looking at the eyes and the ears are they at the same level and while we're here we're going to go ahead and look at the facial expressions and test cranial nerve 7 which is the facial nerve so can you close your eyes tightly for me and open them up okay now smile for me frown and puff out your cheeks okay and he did that with e so that cranial nerve is intact next what we're going to do is we're going to palpate the head the cranium we're going to check for any masses indentations look for skin breakdown any infestations and for this part i like to wear gloves so let's look at the hair so what we're doing is we're filling for any masses indentations and also with this we're looking for any skin breakdown and if your patient's immobile you really want to check the back of the head back here because they're laying on it a lot and there can be breakdown back there also while you're doing that look inside the hair make sure there is no infestations like lice and there's no abrupt like rounding areas of baldness which could represent alopecia then after that since this patient has a beard you want to check the beard as well any lesions any infestations or anything like that and just look around and then once you're done with that what you want to do is you'll doff your gloves and perform hand hygiene next what we're going to do is we're going to find the temporal artery and we're going to palpate them bilaterally and they are both found right here and that his are about a 2 plus and then while we're right there we're gonna go ahead and test cranial nerve five which is the trigeminal nerve and this nerve is responsible for many things like mastication so what i'm gonna have you do then is i'm gonna have you clench your teeth like bite down for me and i'm going to feel the machitar muscle which is right there that should be a nice firm ball and then feel the temporal muscle now what i'm going to do to also test that nerve is have him try to open his mouth against resistance so try to do that for me okay and he can do that now while we're here we're going to go ahead and feel the temporal mandibular joint and we're going to feel right here and i'm going to have you open and close your mouth and i'm feeling for any grading or clicking sensations and i feel none then we're going to palpate the sinuses and i'm going to put pressure on these two sinuses right here and you tell me if you feel any pain okay so the max max maxillary and the frontal no next we're moving down to the eyes and we're going to inspect the eyes first and we're looking at several things we're looking at the eyelid we're looking at the sclera which is the white of the eyes we're looking at the iris we're looking at the pupil and we're looking at the conjunctiva so you shouldn't see any swelling of the eyelids you should see that the sclera is white and shiny it shouldn't be yellow like in jaundice and the conjunctiva when you pull down the lower lid have the patient look up it should be nice and pink it shouldn't be red you shouldn't see any drainage or anything like that and look at the eyes how do they set in the eye socket is are they equal for instance is there any strabismus is there a cross eye where one eye turns in more turns out or up or down and these eyes are normal there's no strabismus next you want to look at anasequaria where you have where one pupil would be smaller than the other people are they equal in size normal pupils should be three to five millimeters in their measurement and here his are about a three and they are equal next what we're going to do is we're going to assess some cranial nerves we're going to be looking at cranial nerve three which is ocular motor four trochular and then six which is abducens and we're gonna do several tests to check their function the first one what we're gonna do is we're gonna be looking for any involuntary shaking of the eye called nystagmus and how we're going to do that is we're going to take our pen light we're going to hold it about 12 to 14 inches away from the patient's nose and then what i want you to do is keep your head still don't move your head and just use your eyes to watch where i move the pin line and as you're doing this you're going to do you're going to perform it in the six cardinal fields of gaze and you're just going to move it and you're looking for any involuntary shaking of the eyes so here we go next we're going to see how reactive the pupils are to lie and to do that we're going to dim the lights a little bit and we're going to have the patient stare off at a distant object that helps dilate those pupils and then we're going to shine using our pin light in at the side and we're going to see how that pupil responds it should constrict and then on the other side it should constrict as well so say their baseline people size was like three millimeters it should go down to one milliliter and it should happen on both sides okay so ben stare off at that object right on the wall over there for me okay and that dilates the pupils and we're just gonna shine light in at this side okay constrict constrict okay let them dilate again then go back to the other side do the same again and they both constricted an equal size next what we're going to do is we're going to check for accommodation and how we do that is we turn the lights back on we just previously had them dim but we now make it light again we're going to have him stare off at a distant object that helps dilate the pupils and we're going to take a pen light you can use a pin light finger and you're just going to slowly move it inward to the nose and what you're looking for is that those pupils constrict they accommodate and the eyes cross while looking at the pen line so here we go stare off in the distance please and i don't want you to move your head or anything just keep it real still and just follow this pin light okay ready okay so now we can document because we just checked all of the things with the eyes we can document that the pupils are equal round reactive to lie and accommodate so that's where that acronym p-e-r-r-l-a comes into play next we're going to move on to the ears so first what we do is we inspect the ears we look on the outside of the ear is there any abnormalities any redness any drainage anything like that and then are you having any pain in your ear okay and sometimes if you have patients who've had long-term gout on the helix of the ear they may have what's called a tophi which is an accumulation of like a whitish yellowish uric acid crystal on the skin so if you ever see that that is what that looks like next we're going to palpate on the ear we're just going to move it around and be and tell me if you have any tenderness whenever i do that and any feel any abnormal masses or lesions and then move the targets a little bit does that hurt or anything like that okay so no pain or tenderness then we're going to palpate the mastoid process which is the big hump behind the ear and we're looking at it is it swollen is there any redness and whenever i touch on it then does it hurt okay and just see if the patient reports any tenderness with that then while you're there you can use the otoscope to inspect the tympanic membrane and remember the tympanic membrane should be a pearly gray translucent color and should be shiny so for an adult you're going to pull the pin of the ear up and back and we're just going to inspect it and also while we're looking at that we're looking at the cone of light and remember the cone of light in the right ear should be at five o'clock and then the left ear should be at seven o'clock next we're going to do one more thing with the ear we're going to test cranial nerve eight which is the vestibulo cochlear nerve and what i'm going to do is i'm going to include one of his ears and then whisper two words on the other side he needs to tell me what i said so you ready okay i'm gonna include this one apple banana okay very good dog cat dog okay and that nerve is intact next we're going to move on to the nose and we're going to inspect the nose we're going to make sure it's midline on the face which it is we're going to look at the septum is it deviated anything like that and ask the patient are you having any trouble with your nose are you having any drainage or anything like that no and you want them to make you want to check the patency of the nose so i mean i'm going to have you occlude one side the nostril breathe out the other and vice versa okay heard airflow airflow nice and patent because sometimes people can have polyps that can block it or the deviated septum then you want to take your pen light and you just want to look inside the nose look for any drainage redness or any like polyps or anything like that and everything looks clear i don't see anything and then we're going to test the olaf factory cranial nerve one the sense of smell so ben what i'm going to have you do is i'm going to have you close your eyes and i want to put something in front of your nose and have you breathe in and smell and you tell me what you smell and whenever you do this use something that's pleasant smelling not something that's really stinky because it could elicit like a gag reflex or something like that if the person has a sensitive nose okay vanilla okay and this was vanilla extract and that's correct so that cranial nerve is intact next we're going to move on to the mouth and for this part like to wear gloves and if your patient is coughing and hacking you might want to wear a mask with a shield so you don't get any mucus on your face or in your mucous membrane so first what we're going to do we're just inspecting the lips make sure they're nice pink color they're not chapped there's no sores on them and one thing with a lot of patients whenever their oxygen saturations are low their lips may turn dusky or blue color so you want to make sure they're nice and pink because that can represent our oxygen level now let's inspect the inside of the mouth but first let's test cranial nerve 12 which is the hypoglossal nerve and what i'm going to have you do then is i'm gonna have you stick out your tongue and move it side to side okay and he does that with ease now what we're gonna do is we're going to inspect the inside of the mouth you'll need a tongue blade for that and just open up your mouth for me and i'm gonna look on the inside of the cheeks nice and pink don't see any sores you're looking to see if they're nice and pink and there's no lesions or anything like that and stick out your tongue for me the tongue should be moist like this and pink you don't want to be beefy red which is like in pernicious anemia you don't want it to be dry or cracked that could be dehydration coat you can put the tongue in then i want you to lift up your tongue for me and look for any lesions underneath the tongue that's where mouth cancer can hang out and i don't see any okay you can close then um you'll while you're also looking at the gums open up a little bit you're gonna look around for cavities any loose or broken teeth no dental carries in there then okay sort of open up your mouth a little bit more put your tongue down and you're going to look at the soft and hard palate now while you're in there you want to look at the uvula make sure it is nice and midline and his is nice and midline and we're going to test cranial nerve 9 the glossopharyngeal so what i'm going to do is i'm going to have you say ah and what you want is that uvula to move up okay and then we're just going to test the gag reflex i'm sort of just going to poke a little bit back there and elicit a gag reply okay there you go kag's really good and um cranial nerve 10 the vegas is intact because he's able to talk with talk to me without hoarseness and he's able to swallow then when you're done inspecting the mouth be sure you take off your gloves and perform hand hygiene now moving on to the neck so what we're going to do is we're going to inspect the neck first so you're going to have the patient extend the neck up a little bit and you're looking at that trachea is it midline look for any lesions and look for any lumps like what you might see in thyroid problems like a gorter and we don't see any of that then what we're going to do is we're going to test cranial nerve 11 which is the accessory nerve so again what i'm going to have you do is move your head side to side up and down okay and then shrug try to shrug against my resistance and he does that with ease so that nerve is intact then we're going to place him at a 45 degree angle and we're going to have him turn his head to the side and what we're looking at is the jugular vein we're looking for any jugular vein distension jvd so again i'm going to just turn your head to the side like that and we're looking for any distinction of the vein and we do not see any next what we're going to do is we're going to palpate so we're going to palpate that trachea just to confirm it is midline and ben do you feel any tenderness or anything like that ask him if he feels any tenderness and i don't feel any lumps the next what we're going to do is we're going to palpate the lymph nodes all sights of those and as i do this tell me if you feel any tenderness and what i'm feeling for is any hard lumps or anything that may be in flame so what we're going to do turn a little bit this way and there we go we're going to start at the pre auricular which is right in front of the ears then we're going to go to the back of the ears the post auricular then we're going to go to the occipital the parotid jugular digastric then we're going to go to the submandibular and then the submental then we're going to go to the superficial cervical and then we're going to make our way down to the deep cervical chain any tenderness so far then we're going to go to the posterior cervical and then right above the clavicle we're going to go to the supraclavicular and not feel anything and no tenderness next we're going to palpate the carotid artery and this is one artery that you do not palpate bilaterally you do one individually so we're going to fill on this side and you're going to find it next to where the groove of the neck and next to the trachea and his is nice and bounding it's two plus then we're just going to fill on the other side and same strength two plus then lastly what we want to do is we're going to auscultate the carotid artery and you're gonna do one side at a time and you're gonna compare sides and you're going to listen with the bell of your stethoscope and we're listening for a brewey which is a swooshing sound so ben what i'm going to do is i'm going to have you breathe in breathe out and hold it for me okay go breathe and breathe out okay you can breathe normally now did not hear it on that side okay breathe in breathe out for me and hold it okay and i did not hear a brewery on that side as well now let's move to the upper extremities so what we're going to do is we're going to inspect the extremities and we're looking for any lesions any redness swelling and this is a good time if they have a central line an iv that you look at that make sure it's not red does iv need to be changed does that pick line or central line need a dressing change assess that then you can palpate and what we're going to do is we're going to palpate our pulse our radial artery so fill those bilaterally and they are two plus and they're equal then we're going to check capillary refill and to do that we're just going to press down on that nail bed and see how fast it comes back and it's less than two seconds then we're going to check skin target by just pinching the skin and see how fast it goes back and that was good then we're just going to look at the range and the motion of the fingers and the hands look at these joints in the hands do you see anything abnormal like for instance like herbidine or beauchard's nodes which are found in osteoarthritis and acetate are you having any pain in your hands or anything like that no then you can palpate the brachial artery which is found in the bends of the arm and just fill those because that's another pulse site and those are two plus and just as a side note if this was a patient that was getting dialysis and they had an av fistula you would want to palpate that and feel for the thrill make sure that that is present up in that arm wherever their fistula is at then you want to test the muscle strength so what we're going to do is i'm going to have you squeeze my fingers as hard as you can okay okay that's really good then i'm gonna have you push up against my hands and i'm gonna push up against your arms okay push okay very good one pin okay and five plus normal strength then we're just going to test his put your hand underneath the elbow and just feel as you move the arm do you feel any grading crepitus of those joints a lot of times in arthritis you can feel that and move that bilaterally another thing you want to do with the upper extremities is to check for a drift and what you will do is you'll have the patient hold out their arms and close their eyes hold it up for about 10 seconds and you're looking for a drift like this so go ahead do that and close your eyes okay and we're assessing to see if this hand will drift upward and a lot of times if a patient has had a stroke okay you can put them down has had a stroke or something like that you will see a drift next we're moving on to the chest and we're going to inspect the chest we're looking for any abnormalities like lesions or any wounds anything like that we're also inspecting the patient's effort of breathing is it really labored are they using those accessory muscles to breathe also we're looking at that anterior posterior diameter so turn to the side like that and you're looking for that barrel chest and it will be increased in patients with like copd they will have what's called the barrel chest and now what we're going to do is we're going to listen to heart sounds and then we're going to listen to lung sounds so first let's auscultate heart sounds and we're going to do this in five locations and they're based on where the valves are located and i like to remember the mnemonic all patients effectively take medicine and the first letter of each word represents the valve except for effectively so a would be aortic p in patients would be pulmonic effectively would be herb's point and this is just the halfway point between the base of the heart and the apex of the heart and there's no valve location there and then t is for tricuspid and then m is for medicine so again using the diaphragm we're going to listen at the right of the sternal border at the second intercostal space and that's going to be the aortic valve so to find that second intercostal space find the sternal notch go down to the angle of lewis then just go a little bit to the right and you're in the second intercostal space and this will be the aortic and we're just listening lub dub lub dub s1 s2 and s2 the dub is going to be louder in this location then we're going to go a little bit over to where the pulmonic valve is found that's on the left of the sternal border at the second intercostal space so just right across again just listening to love dub lub dub and s2 dub is going to be louder in this location then we're going to go a little bit down to the third intercostal space and this is herb's point and again you can hear love dub but there's no specific valve here then we're going to go down to the fourth intercostal space and this is where the tricuspid valve is and lub s1 is going to be the loudest at this location then we're going to go to the fifth intercostal space mid clavicular line and we're going to listen to the mitral valve and again s1 is going to be loudest here dub and there's something special about this site this is the point of maximal impulse this is where you're going to listen for the apical pulse so we're going to set here we're going to count it for one full minute and a normal apical pulse in adult should be 60 to 100 beats per minute and his applicable pulse was 63. then we're going to switch over with the bell of our stethoscope and we're just going to repeat in those locations and we're specifically listening for heart murmurs so that's swishing blowing sound so that's what we're going to listen to with that and i did not hear any now let's listen to lung sounds now when you're listening to lung sounds you're listening for abnormal sounds and here are some samples of some abnormal sounds that you may hear crackles wheezes [Music] a friction rub or strider first we're going to listen anteriorly and what we're going to do is we're going to listen with the diaphragm of our stethoscope and we're going to start at the apex of the lungs and we're going to always compare sides and just enter way downward and assess all the lobes of that right and left lung so first let's start up here okay and then i want you to take good deep breath in and out so here we go apex okay we're going to compare sides then we're going to move down to the second intercostal space and this is going to help us assess the right upper lobe and the left upper lobe so another deep breath in and out then we're going to go down to the fourth intercostal space and we're going to assess where our right middle lobe is and our left upper low because remember the right lung has three lobes and the left lung has two lobes so let's listen to our left upper lobe we're just going to go down a little bit more then we're going to go mid axillary at the sick six intercostal space and we're going to listen to the right and left lower lobe so you just want to turn to the side right there until you get deep breath in for me okay other side okay now let's listen posteriorly again using the diaphragm and the stethoscope you're going to start listening at the apex and work your way down and one thing to keep in mind when you're listening back here you have the scapula and you don't want to listen over those because you won't be able to hear the sound so you're going to listen in between where the scapula and the spine are so down in these regions right here and again we're just going to compare sides and you can do this part at the end if you wanted to whenever you turn your patient over to look at their back side but we're just going to go ahead and do it now so we're going to start here in the apex compare sides then we're going to find c7 which is that vertebral prominence it's the big ball right there you can't miss it and go down to about t3 and you'll be in between the shoulder blades and go a little bit in between the shoulder blades and the spine right in there and you're going to assess the right and left upper lobes then from t3 to t10 we're just going to inch around and we're going to listen to the right and left lower lobes okay now we're going to assess the abdomen and remember we're switching our sequence and how we assess we're going to do inspection auscultation and then percussion or palpation so we're going to do auscultation second so whenever you're looking and assessing the abdomen have the patient lay on their back and what we're going to do is we're going to inspect the abdomen and first i want to ask ben are you having any stomach issues at all no no okay and when was your last bowel movement yesterday morning yesterday morning and how are you urinating do you have any pain while you're peeing do you have problems starting a stream any discharge anything like that no if it's normal okay and with your male patients you want to ask about that due to prostate enlargement was starting a stream and if he was female i would ask him when his last menstrual period was and also again ask a female patient about urinating and things like that now if the patient had a foley this is the time when you would want to look at the urine inspect the foley and look at that just conglomerate your urinary system and your gi system together okay so we're inspecting the abdomen we're looking at the abdominal contour and this patience is scalphoid it goes in a little bit you can also have flat rounded or protuberant and also we're going to know if there's any pulsations a lot of times in this area right here on thin patients like with ben i can see the aortic pulsation in this patient's right above the umbilicus and looking at the belly button and checking for any masses do we see any hernias or anything like that also if your patient had any wounds you would want to look at that and if they had a peg tube you would want to assess the site make sure it's not red and ask them how it feels and with your ostomies with your ostomies you want to look at the stoma and make sure it is like a rosy pink color it's not a dusky cyanotic color and it's not prolapse and look and see what type of stool it's putting out and note that note the smell note when if the bag needs to be changed anything like that so now we're ready to listen to the bowel sounds and what we're going to do is we're going to listen with the diaphragm of our stethoscope and we're going to start in the right lower quadrant and work our way clockwise and we're going to listen all four quadrants and you should hear 5-30 sounds per minute and if you don't hear any bowel sounds you need to listen for five full minutes and you need to note are these normal are they hyperactive or hypoactive so let's listen okay this is our right lower quadrant we're going to move up to the right upper quadrant move over to the left upper quadrant and then down to the left lower quadrant and bowel sounds are normal now we're going to listen for vascular sounds and you're going to do this with the bell of your stethoscope and we're going to listen at the aortic we're going to listen at the renal arteries iliac arteries and you could listen at the femoral artery arteries if you needed to so you're going to listen at the aorta artery and it's a little bit below the xiphoid process a little bit above the umbilicus so about right here and we're listening for like a blowing swishing sound which would represent a brewee okay and none is noted then we're going to listen at the right and left renal arteries which is a little bit down from the aorta location so here's the right okay none note it and then over the left then we're going to listen at the iliac and it's a little bit below the belly button right here and this is the iliac artery and then listen on the other side and again like i pointed out you could listen at the femoral artery in the groin if you needed to now we're going to do palpation first we're going to do light palpation then deep and then as i do this please tell me if you feel any pain or tenderness so first we're going to do light palpation we'll just start in the right lower quadrant work our way around and you're going to go about two centimeters and you're just feeling for any rigidity any lumps masses anything like that how's that feel it feels fine okay okay now we're going to do deep palpation and we're going to go about four to five centimeters so a lot more deep and again you're just feeling for any masses lumps and bend tell me if you have any tenderness and sometimes you can do this with two hands if need be if you're not strong enough like me feeling anything feels nice and soft heard some belly sounds that's why you do this after you listen because you stimulate it okay everything felt good now we're going to assess the lower extremities so first what we're going to do is we're going to inspect we're going to look at the color from the legs to the toes making sure it's nice and pink and here we see that ben has a little bit of a tan line here and we're looking at the hair growth as well you want to make sure there's normal hair growth because in pvd you will see hairless shiny thin legs and here we have excellent hair growth and also do you see any abnormal swelling just right off the bat before you've even touched the patient and look at the legs and the feet for any swelling redness swelling do you have any pain or anything in your legs anything like that and looking at the joints make sure there's no redness on the joints because a lot of times with gout it likes to start out in the big toe so make sure that everything looks good and then on your diabetic patients make sure you look at the bottoms of their feet because these patients don't have the best feeling in their feet so their shoes could be wearing on them or they could have stepped on something and not even know it so inspect those feet make sure there's no ulcers or anything that like that that needs to be addressed also look at the toenails do the toenails look healthy or is there fungus or they missing toenails they have a really bad ingrown toenail so assess for that next you want to palpate your pulses we'll palpate the popliteal pulses which are behind the knee and those are about two plus they're equal bilaterally i'm just filling his legs they're nice and warm and i'm going to push over his tibia firmly and i'm seeing if there's any edema so push there and if there is edema a lot of times whenever you push down it's like this hard light type gel it'll just separate and your finger will leave this indention and here we don't have any now we're going to palpate on the feet and we're going to fill on the pulses and i'm going to dawn gloves perform hand hygiene and dawn gloves and we're going to feel on the pulses and the feet we're going to feel on the posterior tibial and 2 plus really good and then we're going to fill the dorsalis pedis which is on top of the foot two plus with that and if you can't ever find these because sometimes these are hard to find in patients you can get a doppler if you have one on your floor next i'm going to check the capillary refill on his toes just like how we did with the fingers by pushing down in less than two seconds check the other one okay now i'm going to have him push against my hands push against my hands ben okay good job now i'm going to have you raise your legs against resistance good job now we're going to check the babinski reflex and you can use your reflex hammer for this and use the end of it or you can use your finger if you don't have that and what we're going to do is we're going to take this up through the ball of the foot and curve it and we're looking for the toes to curl in which would be a negative normal response so let's check that okay okay and that was normal then we're going to dolph our gloves and perform hand hygiene and next we're going to assess the back so whenever you're looking at the back side you're going to look from the head all the way down at the back and you really want to pay attention to any abnormal moles lesions wounds anything like that and assessing for skin breakdown especially on your patients who are mobile so that would really be concentrating on in the backside area on the coccyx because that's where a lot of breakdown happens and on the back of those heels if you couldn't see it whenever you were assessing the feet and also you could if you hadn't already you could listen to the lung sounds as you have the patient over on the back okay so that wraps up the nursing head to toe assessment now please be sure to check out my other videos because i have a lot of nclex review videos to help you study for nclex along with other nursing skill videos career tips and everything you need to succeed in nursing school all the way to becoming a nurse in your profession so thank you so much for watching and please consider subscribing to this youtube channel
Nursing_Skills_Videos	Clonus_Test_Positive_Reflex_Sign_Preeclampsia_Pregnancy_Nursing_Skills.txt	hey everyone it's sarah stanner sorry and calm and in this video you're going to see a demonstration on how to check for clonus in our previous review we talked about preeclampsia and - the things that we stressed in that lecture was that we needed to check for hyperreflexia and clone a specifically ankle clonise because if those two things are present it indicates that our central nervous system is really stressed out and irritable and there's a risk for seizures so let me show you how to check for clonise and what a positive result looks like first of all you want to do is you want to have the patient's setting and dangling their feet then you're gonna take your hand and you're going to support the lower leg that you want to test and you are going to dorsiflex the foot so I pointed upwards quickly and look for the response and then let it go and that's a normal response that would be a negative clonus a positive clonus you do the same thing you're gonna quickly dorsiflex the foot and you're gonna look for its response a positive response would be that the foot starts to bounce or be at least three or more times and that would be considered a positive clonise okay so that is how you check for clonise and don't forget to check out our other reviews in this maternity series
Nursing_Skills_Videos	Crushing_Medications_for_Tube_Feeding_and_Oral_Adminstration_How_to_Crush_Pills_for_Nurses.txt	hey everyone it's Sarah register nurse aan.com and in this video I'm going to demonstrate for you how to crush a medication when you're giving it through a tube feeding or orally so what is the purpose of why we Crush medications well number one if a patient has a tube feeding we need to crush the medication into a fine powder so we can instill it through the tube so it can get to the stomach or the patient just has trouble swallowing big pills they get choked on them so um we have to crush it and either mix it in water or some type of food like applesauce so let me demonstrate for you how to do this through for a tube feeding or orally and then I'm going to follow up with some tips of some things you want to keep in mind whenever you're doing this skill okay first let me demonstrate for you how to crush a medication to give through a tube feeding okay first before you do this you always want to follow your hospital protocols and how much water you will be giving before you give a medication in between medications and after medications because um each hospital has a strict protocol on what you follow for tube feedings for Med for medication administration and um the following supplies you will need you will need a pill crusher whatever you have access to you will need a medicine cup because you will um be putting the medication in that after you crush it into a fine powder you'll need a 60cc syringe interal syringe that will connect to the patient's tube so you can easily give the medication and um you'll need warm water per peel about 20 CC of warm water that you'll mix each medication in to give and then you'll need a pair of gloves and of course before you do this you do hand hygiene and always put your gloves on whenever you are crushing medications because um you're breaking these medications down into where the chemicals are and you do not want to get these on your skin they can be dangerous okay so you're going to get your pill crusher make sure you have the the right medication the right patient check everything off before you um start doing this and for this particular Crusher you just stick the medication in the bottom and we put this on top and then we just twist it and as we do this it starts to crush the medication and you twist it until it can't go anymore and then untwist it and then I just try to get that residue off at the end so you're making sure you get all the medicine in there and I like to just wipe it in there and then empty it into your medicine cup you got to make sure it doesn't stick on the sides especially if you're using one of these kind of twist type Crushers okay then what we're going to do is we are going to instill some water in our medicine cup and mix it up really good and then we'll draw it up and just stir it because you want everything stirred up very well so it doesn't stick in your tube and cause your tube to become clogged because a clogged tube is no fun for you or the patient and um you will crush each pill separately you never Crush medications together because when you crush them together it will break down that protective coating and those chemal will mix together and um affect their ability to work it can decrease their effectiveness and then you're just going to draw up with your syringe medication and sometimes it likes a settle at the bottom so you've really got to stir it to make sure you can get it all up in there okay and I like to get the air out of my syringe slowly push it up there put your cap back on and then you're ready to instill that into the tube and you'll give that and then flush behind it and then repeat your next medication and if you were using a type of Crusher like this you would need to clean all the residue off wash it with warm soap and water before you do the next medication cuz you don't want to mix medications and that goes for if you're going to be using this for another patient now let me show you how to crush medicines to give orally okay um with this what you can do is you'll mix the medication with either water or here we're going to use applesauce a tip is not to use a patient favorite food or something on their food tray that they're going to be eating like green beans or something like that because these medications when crushed taste horrible and you don't want to ruin the patient's food so um what you're going to do is you need to get your pel Crusher you need to get some applesauce if that's what you're going to use to mix it in a spoon get the right peel for the right patient and then your gloves and of course before you do this perform hand hygiene and put your gloves on and put your pill in your pill crusher and use your pill crusher according to how you're supposed to use it and again you crush pills individually because you don't want to mix them together because they can affect how they work and typically what I like to do is I like to use one spoonful of applesauce per pill and then just repeat with however many pills I have because you don't want to mix one pill and a huge thing of applesauce cuz one cuz if you have to give three more pills that's a lot of applesauce that patients going to be eating so try to use the smallest amount of applesauce as possible so you don't fill your patient up so we're just going to put it in there and be careful you get all of the contents in there okay and remember you will clean your pill crusher in between medications and in between patients if um you're not using a close system Crusher and you will mix this up try to mix it up as good as you can so um patients aren't getting big chunks of pills in their mouth and once you have it mixed good administer that to the patient and then repeat as necessary now let me give you some quick tips to remember whenever you're crushing medications tip number one remember that not all medications can be crushed and if you're ever unsure if a medication can be crushed or not always call Pharmacy because they're great um in helping you determine if you can crush it or not and as a nurse whenever you work over time you get experience you will learn what you can and can't crush and another thing is to look at the medications name because a lot of medications they will have the name and then they'll have these letters after the name like detrol la la stands for long acting you can't Crush that and there's typically five categories of drugs that you cannot crush and they are inter coded long acting extended release controlled release SL delivery or sustained release SL action and to help you remember those five categories of drugs remember this pneumonic seniors erroneously Crush inter coded laxatives and this will help you to remember it because a lot of times patients who are taking Crush medications are your seniors your elderly patients and seniors the ration for seniors is Sr so that's sustained release um erroneously e is for extended release Crush CR is for controlled release and then anic cated is for anic cated and then laxatives is for La which is long acting and the last tip is to try to consult with the physician or the pharmacist on trying to get the medication switched to a liquid form or an IV form so the medication doesn't have to be crushed and um that may be possible if your patient just doesn't like swallowing their pills and they rather have it in a liquid form but if your patient requires like nectar thick liquids or honey thick liquids that they may not be a candidate but always just try to consult with uh other parts of the team to see if you can help the patient out okay so that wraps up this demonstration on how to crush medications thank you so much for watching and don't forget for get to subscribe to our channel for more videos
Nursing_Skills_Videos	Lung_Assessment_Nursing_Lung_Auscultation_Assessing_Lung_Sounds_Part_2.txt	hey everyone it's sarah thread sterner sorry and calm and in this video I want to demonstrate for you on a patient how to auscultate the lungs anteriorly and posteriorly in the previous video I went over an in-depth lecture about how to do this along with normal breath sounds and abnormal breath sounds so be sure to check out that video after you watch this so let's get started first I want to start out showing you the auscultation side so I wanted to use these stickers so you can see about where I'm going to go with the stethoscope and let me get over these real fast for you because these are the little landmarks of where you're going we have our right leg over here and our left leg over here and what we're doing is we're assessing the lungs the right lung has three lobes the right up or the right middle and the right lower the left lobe only half of the left lung only has two lobes the left upper and the left lower so whenever you're listening you want to start out at your apex up at the top of the lungs right above the clavicle just slightly above it right here where you see these pink stickers then you're going to go down to the second intercostal space a little bit over from where your breastbone is the sternum and this is where you're gonna start assessing the upper lobes of the lung then you're just gonna inch down a little bit and you wolf this is about the third intercostal space and you're gonna listen here then go down to the fourth intercostal space and this is one of the landmarks for the right middle lobe your fourth intercostal space so right here well listen I'm here here a right middle lobe and then we'll listen to our left upper lobe then we'll inch down a little bit and we'll listen a little bit more to that right middle lobe and then we're going to be listening to our left lobe and then we're gonna go mid-axillary over here to the side where the armpit is mid axillary axillary sixth intercostal space and we're going to be assessing our lower lobe so it's right here on him and then it's again right here and you'll listen there and then um you'll just inch down a little bit and listen at the bottom just to assess those lower lobes a little bit now let's listen to the lungs with a stethoscope okay before you begin what you want to do is you want to perform here hi Dean introduce yourself to the patient and tell them what you're going to be doing and what whenever I listen to lung sounds I like to listen directly on the skin with my stethoscope not over the gown or over their clothing because you can pick up other sounds that may sound adventitious when it's really not and you just get the best acoustic sound whenever you listen directly on the patient so you want to get your stethoscope and you want to listen with the diaphragm of your set the scope because that's what picks up the best sounds and what we're listening for is we're listening for that full cycle of inspiration and expiration and we're paying attention to what type of breath sound is it is it bronchial vesicular Branko vesicular if you're not familiar with those sounds check out my other videos because I have audio clips where you can listen to those and we're also listening for any extra sounds on top of those sounds like crackles wheezes Strider anything like that that we may be picking up and another kids this skill takes getting used to you have to practice it over and over especially when you're listening over these areas because you're gonna hear the heartbeat at the same time so you got to train your ears to listen for those correct sounds like you're the heart versus lungs so practice this on yourself and others so you can get comfortable with it so what we're gonna do is we're gonna start at the top and we're gonna work our way down and we're going to start on one side and then compare that side because we want to say say see if we're hearing the same thing on this side as we are that side so let's get started okay what I'm gonna have you do is take one breath in through your mouth and out through your mouth and if you start to get dizzy or anything let me know and we can slow down okay we're starting up at the apex right above the clavicle now we're gonna inch down to the second intercostal space and this is where we're gonna start listening to our upper lobes of our lungs so about right there then we're just gonna inch down a little bit to about the third intercostal space and listen there and we're gonna go over and compare it now we're going to go down into the fourth intercostal space and we're still gonna be in the upper lobe of the left lung but now we're gonna be in the middle lobe of the right lung so let's just go down just a little bit and then we're gonna go over here and compare and then we'll just go a little bit do maybe the fifth intercostal space just still assessing that middle lobe and I go over here and compare now we're gonna go mid axillary and we're gonna be down in our lower lobes now and this will be at the six intercostal space okay we're going to compare sides so if you just want to turn over here we're gonna go mid acts like a parasite then we're just going to go down just a little bit to say in that lower lobe and then we're going to compare sides now let's look at the posterior view of the back here over on this side we have the left lung and over here we have the right lung it's flip compared to the anterior view and what we'll do is we'll start listening up at the apex of the lungs again and that's right above slightly above that shoulder blade in this area where the green stickers are then we're going to go and we're gonna listen between c7 to t3 and c7 is where you're prominent starts so right here and we'll go a little bit below it because c7 to t3 is going to assess the upper lobes of the posterior side of the lungs so we'll go here and here then we'll go down just bed.this is about where t3 is and then we'll jump down to round t4 and from t3 to t10 we'll be assessing the lower lobes in your posterior view you have more ability to auscultate the lower lobes then compared to the anterior view where you're gonna really be assessing the upper lips so then we'll just go down listen here listen there we'll go there and then we'll just go a little bit mid-axillary almost mid-axillary and listen over on these sides and then down at the bottom and over here now let's do it with a stethoscope now we're going to listen posterior lis okay let me tell you a trick with this whenever you're listening back here as you've learned if you watch the other lectures you have the scapula back here and you have the vertebrae and what you're trying to do is to get in between that scapula and vertebra to listen to those lungs and you don't want to go over the shoulder blades of scapula while you're listening to lungs because it's really gonna muffle out your sound and you're not gonna hear a lot so what you need to do is have the patient sort of just move their arms like this to separate those shoulder blades from one another so you have some space to work with so what we're gonna do again we're gonna start from top and work our way to the bottom and we're going to compare side so we're gonna go slightly above the scapula listen to the apex then we're gonna go down in the area where c7 is which is where your prominence is then we'll work our way down to where t3 and remember this is assessing the upper lobe and then we'll go down from t3 to about t10 and assess those lower lobes so here we go and again just take a deep breath for me in and out okay and we are done with posterior okay so that is how you assess the lungs with a stethoscope now be sure to check out the other videos in this lung series like the audio files of the normal breast films vs. abnormal and the lecture that covers this material in depth and thank you so much for watching and please consider subscribing to this YouTube channel
Nursing_Skills_Videos	Nitropaste_Ointment_Application_Nitroglycerin_Nitro_Bid_Medication_Administration_Nursing.txt	hey everyone it's sarah stanner sorry end comm and in this video i'm gonna demonstrate for you how to administer nitro bid which is not your glycerine appointment first what you want to do is you want to gather your supplies so you'll need to get your nitro bid nitro glycerine pointment and a lot of them come in tubes like this or they come in like little foil packages so get whatever your facility has you will definitely need to get an application paper this is so important because this allows you to measure how many inches of ointment you're going to get like half an inch an inch and they come in like a little booklet like this you can tear off in your Omni cell whatever you have that dispenses medication and get one of these papers you also need to get some gloves to protect yourself from this ointment because if you get nitroglycerin on you this ointment it will cause a headache and cause you to feel flushing and things like that and you don't want to experience that you need some tape to secure down the application papers along with some Tegaderm or plastic wrap to cover the application site because nitroglycerin can stain the clothing along with a cloth for whenever you remove the old application paper from the skin just to clean off that excessive amount of nitroglycerin that will be left behind after you gather your supplies you'll want to do the patient's five right so make sure you have the right patient you have the right drug the right dose the right time and the right route now whenever you're giving nitroglycerin appointment you want to watch out for some things number one you want to make sure the patient's blood pressure is good they don't have hypotension and you want to warn your patient that you know this may cause a headache if this is really the first time receiving it it can also cause warmness and flushing in the space another thing is you need to know what nitroglycerin pointment is used for it's typically used to prevent chest pain in patients with coronary artery disease so if you're working on a cardiac floor you'll probably be giving a lot of nitro Paste so the first thing what you want to do is you want to remove the old application paper first that contains the nitroglycerin before we actually put on the new one so you want to wash your hands and you want to wear gloves very important never do this without gloves because there's still nitroglycerin on the skin and if you get this on you you will get some side effects of nitroglycerin and you don't want that so we are going to take it off and gently just remove the tape from the site and if you have trouble finding where the the applicator pad is just ask the patient and just look everywhere you can find them on the arm on the chest on the back sometimes maybe the patient's bed if they've came off so just always try to look for it and then you're just gonna take it off notice there's still some nitro and then just gently fold this together and discard then take your cloth and gently just wipe off the area of the excessive nitroglycerin now we're going to place a new applicator pad somewhere else and again we can use the arm the chest the back but you always want to rotate sites after doing that you want to perform hand hygiene then you want to prep your application paper and what we're gonna do is we're going to date and time it and initial it and then we're going to square the ointment on to the scale so where we're going to write our date and time and initials is on the printed side because on this opposite side without the print is where we are going to measure the pointment and you can see through the onto the other side because it's paper so thin so you can measure it a lot of people ask well why don't we just measure it on the side with the scale it's because whenever the pointment comes into contact with this print it can cause the print to leak into the ointment and we don't want to get that on the patient so we're gonna go ahead and date time and initial it and don't date in time over the scale because when you flip it over you won't be able to see the scale whenever you're measuring out the pointman so right somewhere where you have access to it and it's best to use one of these markers if you can because it shows up better than a pen okay now we're going to flip it over and we're going to measure out our point Minh and the doctors order says one inch of Nitra bid point Minh so to do that we are going to daughter gloves now let's open our pointment tubing okay if this is the first time you're opening up this point mandir is a special way you have to open it and for this video this is a demo dose of nitroglycerin pointment it's not the real thing in there we're just using this for teaching purposes so what you want to do is you want to unscrew this cap and on this side of the cab is like this beveled area that's sort of pointy it's going to screw into the top of this tubing and it'll poke a hole in it so you can squirt out the ointment so take that and just insert it in there and just screw it on a little bit like that and see it allows your ointment to come out now we're going to squirt out one inch of paste okay so looking at your scale from here to here is half an inch and from here to here is an inch and so we're going to squirt the ointment from here to there so let's do that okay and we have an inch so we picked the other arm to place our new application paper with appointment and I made sure that it was intact there's no areas that are open and there it's free from hair if it had a lot of hair we want to trim it not shave it and I wanted to make sure that the site was clean so now we're just going to apply our paper to it and to do that we are going to put the ointment side on to the skin and we're just going to thinly spread it at the side don't rub it in don't massage it just thinly spread it so it's evenly dispersed like that now we're going to secure it to keep it in place so you can use a Tegaderm plastic wrap whatever you have access to and just cover the site and then tape it into place okay and then the patient is ready to go until their next dose after you're done with that discards your gloves perform hand hygiene and then chart that you gave them medication and where you place that applicator pad so the next nurse can find it and just a quick tip I wanted to go over real quick with you is about this nitro pointment and some places keep it in the tubing like here and if this is the case with your facility because it's multi patient use you can use it for your patient several times in the tube you'll put like their patient label on it you want to keep this in the designated area that your hospital wants you to keep it not in the patient's room and their drawer or something like that because one time this nurse told me that she had a patient and she walked into the room and the wife was giving the patient a massage a back massage and she was using this ointment because she found the women in the patient's drawer and thought it was just some type of lotion she could use and both her both the patient and the patient's wife had an adverse reaction to it so just to protect your patient their family members and yourself always keep this in the designated place set by your Hospital okay so that wraps up how to administer nitroglycerin thank you so much for watching and don't forget to subscribe to our channel for more videos
Nursing_Skills_Videos	Visual_Acuity_Test_with_Snellen_Eye_Chart_Exam_Cranial_Nerve_2_Assessment_Nursing.txt	this is cereth is Turner's orange calm and in this video I'm going to show you how to assess visual acuity using a Snellen chart whenever you do this you're assessing cranial nerve 2 which is the optic nerve and of course whenever you perform the skill you will need a Snellen chart now we're going to test visual acuity using a Snellen chart and what you're gonna do is you're gonna have your patients and about 20 feet from the chart so Ben if you'll stand about right there for me and ask your patient do you wear glasses ok and if your patient does wear glasses you'll want them to wear those for this test ok so what we're gonna do look at that chart over there and try to read the lowest line for me that you can read ok and first we're gonna cover your right eye then your left eye and then we'll do both eyes ok so cover your right eye ok and what line can you read ok read it for me ok very good ok now we're going to cover up your left eye and do the same thing and again whatever line you can read let me know 8 again ok the C ok and now read with both eyes and what line ok ok and he read from line 8 so that means that he has 20/20 vision and this means that he can see the same line of letters at 20 feet that a person with normal vision can see at 20 feet however let's say that in his left eye he could only read like line 6 which is 2030 that would mean that his left eye sees at 20 feet that a person with normal vision would see at 30 okay so that is how you assess visual using a Snellen chart thank you so much for watching and don't forget to subscribe to our channel for more videos
MIT_6034_Artificial_Intelligence_Fall_2010	15_Learning_Near_Misses_Felicity_Conditions.txt	PROFESSOR PATRICK WINSTON: You know, some of you who for instance-- I don't know, Sonya, Krishna, Shoshana-- some of you I can count on being here every time. Some of you show up once in a while. The ones of you who show up once in a while happen to be very lucky if you picked today, because what we're going to do today is I'm going to tell you stuff that might make a big difference in your whole life. Because I'm going to tell you how you can make yourself smarter. No kidding. And I'm also going to tell you how you can package your ideas so you'll be the one that's picked instead of some other slug. So that's what we're going to do today. It's the most important lecture of the semester. The sleep lecture is only the second most important. This is the most important. Now the vehicle that's going to get us there is a discussion about how it's possible to learn in a way that is a little reminiscent of what we talked about last time. Because last time we learned something very definite from a small number of examples. This takes it one step further and shows how it's possible to learn in a human-like way from a single example in one shot. So it's extremely different, very different from everything you've seen before. Everything that involves learning from thousands of trials and gazillions of examples and only learning a little tiny bit, if anything, from each of them. This is going to learn something definite from every example. So here's the classroom example. What's this? It's an arch. I know the architects are complaining that it's not an arch in architecture land. It's a post and lintel construction. But for us today it's going to be an arch. Now if you were from Mars and didn't know what an arch was, I might present this to you and you'd get a general idea of some things that might be factors, but you'd have no idea what's really important. So then I would say, that's not an arch. And you would learn something very definite from that. And then I would shove these together and put this back on, and I would say, that's not an arch either. And you'd learn something very definite from that. And then I could paint the top one blue, and you'd learn something very different from that. And how can that happen is the question? How can that happen in detail, and what might it mean for human learning and how you can make yourself smarter? And that's where we're going to go. All right? So how can we make a program that's a smart as a martian about learning things like that? Well, if you were writing that program, surely the first thing you would do is you'd try to get off the picture as quickly as possible and into symbol land where things are clearer about what the important parts are. So you'd be presented with an initial example that might look like this. We'll call that an example. And it's more than just an example. It's the initial model. That's the starting point. And now we're going to couple that with something that's not actually an arch but looks a whole lot like one, at least on the descriptive level to which we're about to go. So here's something that's not an arch, but its description doesn't differ from that of an arch very much. In fact, if we were to draw this out in a kind of network, we would have a description that looks like this, and these relations would be support relations. And this would be drawn out like so. And the only difference would be-- the only difference would be that those support relations that we had in the initial model-- the example-- have disappeared down out here in this configuration. But since it's not very different from the model, we're going to call this a near miss. And now, you see, we've abstracted away from all the details that don't matter to us. Last time we talked about a good representation having certain qualities-- qualities like making the right things explicit. Well, this makes the structure explicit, and it suppresses information about blemishes on the surface. We don't care much about how tall the objects are. We don't think it matters what they're made of. So this is a representation that satisfies the first of the criteria from last time. It makes the right things explicit. And by making the right things explicit, it's exposing some constraint here with respect to what it takes to be an arch. And we see that if those support relations are missing, it's not an arch. So we ought to be able to learn something from that. What we're going to do is we're going to put these two things together. We're going to describe the difference between the two. And we're going to reach the conclusion that since there's only one difference-- one kind of difference with two manifestations to disappearing support relations, we're going to conclude that those support relations are important. And we're going to turn them red because they're so important. And we're going to change the name from "support" to "must support." So this is our new model. This is an evolving model that now is decorated with information about what's important. So if you're going to match something against this model, it must be the case that those support relations are there. If it's not there-- if they're not there, it's not an arch. All right? So we've learned something definite from a single example. This is not 10,000 trials. This is a teacher presenting something to the student and the student learning something immediately in one step about what's important in an arch. So let's do it again. That was so much fun. Let's do this one. Same as before except that now when we describe this thing, there are some additional relations-- these relations, and those are touch relations. So now when we compare that-- is that an arch? No. It's a near miss. When we compare that near miss with our evolving model, we see immediately that once again there's exactly one difference, two manifestations, the touch relations. So we can immediately conclude that these touch relations are interfering with our belief that this could be an arch. So what do we do with that? We put those together again and we build ourselves a new model. It's much like the old model. It still has the imperatives up here. We have to have the support relations. But now down here-- and we draw not signs through there-- these are must not touch relations. So now you can't match against that model if those two side supports are touching each other. So in just two steps, we've learned two important things about what has to be in place in order for this thing to be construed to be an arch. So our martian is making great progress. But our martian isn't through, because there's some more things we might want it to know about the nature of arches. For example, we might present it with this one. Well, that looks just like our initial example. It's an example just like our initial example. But this time the top has been painted red. And I'm still saying that that's an arch. So once again, there's only one difference and that difference is that in the description of this object, we have the additional information that the color of the top is red. And we've been carrying along without saying so, that the color of the top in the evolving model is white. So now we know that the top doesn't have to be white. It can be either red or white. So we'll put those two together and we'll get a new model. And that new model this time once again will have three parts. It will have the relations, an imperative form that we've been carrying along now, the must support and the must not touch, but now we're going to turn that color relation itself into an imperative. And we're going to say that the top has to be either red or white. So now, once again, in one step we've learned something definite about archness. Two more steps. Suppose now we present it with this example. It's an example. And this time there's going to be a little paint added here as well. This time we're going to have the top painted blue like so. So the description will be like so. And now we have to somehow put that together with our evolving model to make a new model. And there's some choices here. And our choice depends somewhat on the nature of the world that we're working in. So suppose we're working in flag world. There are only three colors-- red, white, and blue. Now we've seen them all. If we've seen them all, then what we're going to do is we're going to say that the evolving model now is adjusted yet again like so. Oh-- but those are imperatives still. Let me carry that along. At this time, this guy-- the color relation-- goes out here to anything at all. So we could have just not drawn it at all, but then we would have lost track of the fact that we've actually learned that anything can be there. So we're going to retain the relation but have it point to the "anything goes" marker. Well, we're making great progress and I said there's just one more thing to go. So let me compress that into this area here. What I'm going to add this time is I'm going to say that the example is like everything you've seen before except that the top is now one of those kinds of child's bricks. So you have a choice actually about whether this is an arch or not. But if I say, yeah, it's still an arch, then we'd add a little something to its description. So this description would look like this. Same things that we've seen before in terms of support, but now we'd have a relation that says that this top is a wedge. And over here-- something we've been carrying along but not writing down-- this top is a block. A brick, I guess in the language of the day. So if we say that it can be either a wedge or a brick on top, what do we do with that? Once again, it depends on the nature of representation, but if we say that we have a representation, that has a hierarchy of parts. So bricks and wedges are both children's blocks and children's box or toys. Then we can think of drawing in a little bit of that hierarchy right here and saying well, let's see. Immediately above that we've got the brick or wedge. And a little bit above that we've got block. And a little bit above that we've got toy. And a little bit above that we eventually get to any physical object. So what does it do in response to that kind of situation? You have the choice. But what the program I'm speaking of actually did was to make a conservative generalization up here just to say that it's one of those guys. So once again it's learned something definite. Let me see. Let me count the steps. One, two, three, four, five. And I just learned four things. So the generalization of a color, it took two steps to get all the way up to "don't care." So note how it contrasts with anything you've seen in a neural net. Or anything you will see downstream in some of the other learning techniques that we'll be talking about that involve using thousands of samples to learn what it is-- to learn whatever it is that is intended to be learned. Let me show you another example of how these heuristics can be put to work. So there are two sets of drawings. We have the upper set and the lower set. And your task, you smart humans working in vast parallelism, your task is to give me a description of the top trains that distinguishes and separates them from the trains on the bottom. You got it? Nobody's got it? Well, let me try one on you. The top trains all have a short car with a closed top. So how is it possible that a computer could have figured that out? It turns out that it figured it out with much the same apparatus that I've shown you here in connection with the arches, just deployed in a somewhat different manner. In this particular case, the examples are presented one at a time by a teacher who's eager for the student to learn. In this case, the examples are presented all at once and the machine is expected to figure out a description that separates the two groups. And here's how it works. What you do is you start with one of them. But you have a lot of them. You have some examples-- we'll call the examples on top the "plus examples" and the examples on the bottom the "negative examples." So the first thing that you do is you pick one of the positive examples to work with. Anybody got any good guesses about what we're going to call that? Yeah, you do. We're going to call that the seed. It's just highly reminiscent of what we did last time when we were doing [? phonology ?] but now at a much different level. We're going to pick one of those guys to be the seed, and then we're going to take these heuristics and we're going to search for one that loosens this description so that it covers more of the positives. You see, if you have a seed that is exactly a description of a particular thing and you insist that everything be just like that, then nothing will match except itself. But you can use these heuristics to expand the coverage of the description, to loosen it so that it covers more of the positives. So in your first step you might cover, for example, that group of objects. Too bad for your side, you've also in that particular case included a negative example in your description, but perhaps in this next step beyond that you'll get to the point where you've eliminated all of those negative examples and zeroed in on all the positive examples. So how might a program be constructed that would do that sort of thing? Well, think about the choices. The first choice that you have it is to pick a positive example to be the seed. And once you've picked a particular example to be the seed, then you can apply heuristics, all of them that you have, to make a new description that may cover the data better. It may have more of the positives and fewer of the negatives than in your previous step. But this, if you have a lot of heuristics, and these are a lot of heuristics because there's a lot of description in that set of trains, there are lots of possible things that you could do with those heuristics because you could apply them anywhere. So this tree is extremely large. So what do you do to keep it under control? Well, now you have answers to questions like that by knee-jerk, right? The branching factor is too big. You want to keep a few solutions going. You have some way of measuring how well you're doing so you can use a beam search. This piece here was originally worked out by a friend of mine, now, alas, deceased, [? Rashad ?] [? Malkowski ?] when he was at the University of Illinois. And of course, he wasn't interested in toy trains, he was just interested in soybean diseases. And so this exact program was used to build descriptions of soybean diseases. It turned out to be better than the plant pathology books. We now have two ways of deploying the same heuristics. But my vocabulary is in need of enrichment, because I'm talking about "those" heuristics. And one of the nice things that [? Malkowski ?] did for me a long time ago is give each of them a name. So here are the names that were developed by [? Malkowski. ?] What's happening here? You're going from an original model to an understanding-- some things are essential. So he called this the "require link" heuristic. And here in the next step, we're forbidding some things from being there. So [? Malkowski ?] called that heuristic the "forbid link" heuristic. And in the next step, we're saying it can be either red or white. So we have a set of colors and we're extending it. And over here in this heuristic, going from red or white to anything goes, that's essentially forgetting about color altogether, so we're going to call that "drop link" even though for reasons of keeping track, we don't actually get rid of it. We just have it pointing to the "anything" marker. And finally, in this last step, what we're doing with this tree of categories is we're climbing up it one step. So he called that the "climb tree" heuristic. So now we have a vocabulary of things we can do in the learning process, and having that vocabulary gives us power over it, right? Because those are names. We can now say, well, what you need here is the "drop link" heuristic. And what you need over there is the "extend set" heuristic. So now I want to back up yet another time and say, well, let's see. When we were working with that phonology stuff, all I did was generalize. Are we just generalizing here? No. We're both generalizing and specializing. So when I say that the links over here that are developed in our first step are essential, this is a specialization step. And when I say they can't be-- they cannot be touch relations, that's a specialization step. Because we're able to match fewer and fewer things when we say you can't have touch relations. But over here, when I go here and say, well, it doesn't have to be white. It can also be red. That's a generalization. Now we can match more things. And when I drop the link altogether, that's a generalization. And when I climb the tree, that's a generalization. And that's why when I do this notional picture of what happens when [? Malkowski ?] program does a tree search to find a solution to the train problem, they're both specialization steps which draw in the number of things that can be matched, and generalization steps that make it broader. So, let's see. We've also got the notion of near miss. And we've got the notion of example-- some of these things are examples, some are near misses. We've got generalization specialization. Does one go with one or the other, or are they all mixed up in their relationship to each other? Can you generalize and specialize with near misses? What do you think? You think-- you don't think so, [INAUDIBLE]? What do you think? STUDENT: [INAUDIBLE] specialization. PROFESSOR PATRICK WINSTON: [INAUDIBLE] lead to specialization. Let's see if that's right. So we've got specialization here, and that's a near miss. We've got specialization here, and that's a near miss. We've got generalization here, and that's an example. And we've got generalization here, and that's an example. And we've got generalization here, and that's an example. So [INAUDIBLE] has got that one nailed. The examples always generalize, and the near misses always specialize. So we've got apparatuses in place that allow us to both expand what we could match and shrink what we could match. So what has this got to do anything? Well, which one of these methods is better, by the way? This one-- this one requires a teacher to organize everything up. This one can handle it in batch mode. This one is the sort of thing you would need to do with a human because we don't have much memory. That one is the sort of thing that a computer's good at because it has lots of memory. So which one's better? Well, it depends on what you're trying to do. If you're trying to build a machine that analyzes the stock market, you might want to go that way. Or soybean diseases, or any one of a variety of practical problems. If you're trying to model people, then maybe this is a way that deserves additional merit. How do you get all that sorted out? Well, one way to get it all sorted out is to talk in terms of what are sometimes called "felicity conditions." So when I talk about felicity conditions, I'm talking about a teacher and a student and covenants that hold between them. So here's the teacher. That's me. And here's the student. That's you. And the objective of interaction is to transform an initial state of knowledge into a new state of knowledge so that the student is smarter and able to make use of that new knowledge to do things that couldn't be done before by the student. So the student over here has a learner. And he has something that uses what is learned. And the teacher over here has a style. So if any learning is to take place, one side has to know something about the other side. For example, it's helpful if the teacher understands the initial state of the student. And here's one way of thinking about that. You can think of what you know as forming a kind of network. So initially, you don't know anything. But as you learn, you start developing quanta of knowledge. And these quanta of knowledge are all linked together by prerequisite relationships that might indicate how you get from one quantum to another. So maybe you have generalization links, maybe you have specialization links, maybe you have combination links, but you can think of what you know as forming this kind of network. Now your state of knowledge at any particular time can then be viewed as a kind of wavefront in that space. So if I, the teacher, know where your wavefront is, can I do a better job of teaching you stuff? Sure, for this reason. Suppose you make a mistake, m1, that depends on q1. Way, way behind your wavefront. What do I do if I know that you made a mistake of that kind? Oh, I just say, oh, you forgot you need a semicolon after that kind of statement. I just remind you of something that you certainly know, you just overlooked. Right? On the other hand, suppose you make a mistake that depends on a piece of knowledge way out here. That kind of mistake, m2. What do I say to you then? What do you think, Patrick? What do you think I would say if you made that kind of mistake? STUDENT: [INAUDIBLE]. PROFESSOR PATRICK WINSTON: No. That's not what I would say [INAUDIBLE]. STUDENT: You'd tell us that we don't know that yet. PROFESSOR PATRICK WINSTON: I would say something like that. What [INAUDIBLE] suggested I would say. Oh, don't worry about that. We'll get to it. We're not ready for it yet. So in this case, I remind somebody of something they already know. In this case, I tell them they'll learn about it later. So what do I do with mistake number three? That's the learning moment. That's where I can push the wavefront out. Because everything's in place to learn the stuff at the next radius. So if I know that the student has made a mistake on that wavefront, that's when I say, this is the teaching moment. This is when I explain something. So that's why it's important for the teacher to have a good model of where the student is in the initial state of knowledge. Next thing that's important for the teacher to know is the way that the student learns. Because if the student is a computer, they can handle the stuff in batch. That's one thing. If the student is a third grader who has a limited capacity to store stuff, then that makes a difference in how you teach it. You might teach it that way to the third grader, and that way, buried underneath this board, to a computer. So you need to understand the way that the learner-- the computational capacity of the learner. And there's also a need to understand the computational capacity of the user box down there, because sometimes you can be taught stuff that you can't actually use. So by now, most of you have attempted to read that sentence up there, right? And it seems screwy, right? It seems unintelligible, perhaps? It's a garden path sentence. It makes perfectly good English, but the way you generally read it, it doesn't, because you have a limited buffer in your language processor. What does this mean? You're expecting this to be "to." Question. But it's actually a command. Here's the deal. Somebody's got to give the students their grades. Well, we can have their parents do it. Have the grades given to their students by their parents, then. So it's a command. And you garden path on it, because you have limited buffer space in your language processor. So with parentheses you can understand it. You can learn about it. You can see that it's good English, but you can't generally process that kind of sentence without going back and starting over. And what about going the other way? Are there covenants that we have to have here that involve the student understanding some things about the teacher? Well, first thing there is is trust. The student has to presume that the teacher is teaching the student correct information, not lying to student. Ratified that you're all here because presumably you all think that I'm not trying to screw you by telling you stuff that's a lie. There's also this sort of thing down here. Understanding of the teacher's style. So you might say, well, professor x, all he does is read slides to us in class, so why go? You wouldn't be entirely misadvised. That's an understanding of one kind of style. Or you can say, well, old Winston, he tries to tell us something definite and convey a family of powerful ideas in every class. So maybe it's worth dragging yourself out of bed at 10 o'clock in the morning. Those are style issues, and those are things that the student uses to determine how to match the student's style against that of the instructor. So that helps us to interpret or think about differences in style so that we can appreciate whether we ought to be learning that way, where that way is the way that's underneath down here, the way you would teach a computer, the way [? Malkowski ?] taught a computer about soybean diseases. We can do it that way, or we can do it this way with a teacher who deliberately organizes and shapes the learning sequence for the benefit of a student who has a limited processing capability. Now you're humans, right? So think about what the machine has to do here. The machine-- in order to learn anything definite in each of those steps, the machine has to build a description. So it has to describe the examples to itself. That's unquestioned, right? Because what it's doing is looking at the differences. So it can't look at the differences unless it's got descriptions of things. So if you're like the machine, then you can't learn anything unless you build descriptions. Unless you talk to yourself. And if you talk to yourself, you're building the kind of descriptions that make it possible for you to do the learning. And you say to me, I'm an MIT student. I want to see the numbers. So let me show you the numbers. And when I'm going to show numbers-- the numbers that I'm going to show you show you the virtues of talking to yourself. So here's the experiment. The experiment was done by a friend of mine, Michelene Chi. Always seems to go by the name Mickey Chi. There he is. So here's the deal. The students that she worked with were expected to learn about elementary physics. 801 type stuff. And she took eight subjects, and she had them-- she took them through a bunch of examples and then she gave them an examination. So eight subjects, and so they divide into two groups. The bottom half and the top half. The ones who did better than average and the ones who did worse than average. So then you can say, well, OK, what did that mean? You can say, how much did they talk to themselves? Well, that was measured by having them talk out loud as they solved the problems on an examination. So we could ask how much self explanation was done by the smart ones versus the less smart ones? And here are the results. The worst ones-- the worst four said about 10 things to themselves. The best four said about 35 things to themselves. That's a pretty dramatic difference. Here's the data in a more straightforward form. This, by the way, points out that the smart ones scored twice as high as the less smart ones. And when we look at the number of explanations they gave themselves in two categories, smart ones said three times as much stuff to themselves as the less smart ones. So, as you can see, the explanations break down into two groups. Some have to do with monitoring and not with physics at all. They're things like, oh hell, I'm stuck. Or, I don't know what to do. And the others have to do with physics. Things like, well, maybe I should draw a force diagram. Or let me write down f equals ma, or something like that, as physics knowledge. I think it's interesting that this average score is different by a factor of two, and the average talking to oneself differed by a factor of three. Now this isn't quite there, because what's not clear is if you encourage somebody to talk to themself, and they talk to themselves more than they would have ordinarily, does that make them score better? All we know is that the ones who talk to themselves more do score better. But anecdotally, talking to some veterans of 6.034, they've started talking to themselves more when they solve problems, and they think that it makes them smarter. Now I would caution you not to do this too much in public. Because people can get the wrong idea if you talk to yourself too much. But it does seem-- it does, in fact, seem to help. Now what I did last time is I told you how to be a good scientist. What I'm telling you now is how to make yourself smarter. And I want to conclude this hour by telling you about how you can package your ideas so that they have greater impact. So I guess I could have said, how to make yourself more famous, but I've limited myself to saying how to package your ideas better. And the reason you want to package your ideas better is because if you package your ideas better than the next slug, then you're going to get the faculty position and they're not. If you say to me, I'm going to be an entrepreneur, same thing. You're going to get the venture capitalist money and the next slug won't if you package your ideas better. So this little piece of work on the arch business got a whole lot more famous than I ever expected. I did it when I was young and stupid, and didn't have any idea what qualities might emerge from a piece of work that would make it well known. I only figured it out much later. But in retrospect, it has five qualities that you can think about when you're deciding whether your packaging of your idea is in a form that will lead to that idea becoming well known. And since there are five of them, it's convenient to put them all on the points of a star like so. So quality number one. I've made these all into s-words just to make them easier to remember. Quality number one is that there's some kind of symbol associated with a work. Some kind of visual handle that people will use to remember your idea. So what's the visual symbol here? Well, that's astonishingly easy to figure out, right? That's the arch. For years without my intending it, this was called arch learning. So you need a symbol. Then you also need a slogan. That's a kind of verbal handle. It doesn't explain the idea, but it's enough of a handle to, as Minsky would say, put you back in the mental state you were in when you understood the idea in the first place. So what is the slogan for this work? Anybody have any ideas? Pretty obvious. What's essential to this process working? The ability to present an example is very similar [INAUDIBLE], that constitutes a model but isn't one of those. STUDENT: [INAUDIBLE]. PROFESSOR PATRICK WINSTON: So it's a near miss. The next thing you need if your work is going to become well known is a surprise. What's the surprise with this stuff? Well, the surprise-- everything that had been done in artificial intelligence having to do with learning before this time was precursors to neural nets. Thousands of examples to learn anything. So the big surprise was that it was possible for a machine to learn something definite from each of the examples. So that now goes by the name of one shot learning. That was the surprise, that a computer could learn something definite from a single example. So let's see. We've almost completed our star. But there are more points on it. So this point is the salient. What's a salient-- what's a salient idea? Jose, do you know what a salient idea is? He's too shy to tell me. What's a salient idea? Ah, who said important? Wrong answer, but very good. You're not shy. So what does it really mean? Yes. STUDENT: Relative to what somebody's already thinking about? PROFESSOR PATRICK WINSTON: Relative to what somebody's thinking about. Not quite. If you have a-- if you're an expert in-- yes? STUDENT: [INAUDIBLE]. PROFESSOR PATRICK WINSTON: Really close. We're getting closer. [INAUDIBLE]. Yes? STUDENT: Maybe an idea that wasn't obviously apparent, but becomes apparent gradually as somebody starts to understand? PROFESSOR PATRICK WINSTON: We're zeroing-- we're circling the wagons here and zeroing in on it. Yes? STUDENT: If I'm preempting what you're about to say, it has sort of a doorway of how you can understand the idea. PROFESSOR PATRICK WINSTON: It's what? Sorry. STUDENT: It's sort of like a doorway of how you can grasp the idea. PROFESSOR PATRICK WINSTON: That's sort if it, too, but if you study military history, what's the salient on a fort? Well, this is a good word to have in your vocabulary because it sort of means all of those things, but what it really means is something that sticks out. So on a fort, if this were a fort, these would all be salients because they stick out. So the salient idea is usually important because it sticks out. But it's not-- the meaning is not "important," the meaning is "stick out." So a piece of work becomes more famous if it has something that sticks out. It's interesting. There are theses that have been written at MIT that have too many good ideas. And how can have too many good ideas? Well, you can have too many good ideas if no one idea rises above and becomes the idea that people think about when they think about you. We have people on the faculty who would have been more famous if their theses had fewer ideas. It's amazing. So this piece of work did have a salient. And the salient idea was that you could get one shot learning via the use of near misses. That was the salient idea. The fifth thing, ah. Talk more about this in my "How to Speak" lecture in January. The fifth thing I like people to try to incorporate into their presentations is a story. Because we humans somehow love stories. We love people to tell us stories. We love things to be packaged in stories. And believe me, I think all of education is essentially about storytelling and story understanding. So if you want your idea to be sold to the venture capitalist, if you want to get the faculty job, if you want to get your book sold to a publisher, if you want to sell something to a customer, ask yourself if your presentation has these qualities in it. And if it has all of those things, it's a lot more likely to be effective than it doesn't. And you'll end up being famous. Now you say to me, well, being famous-- that sounds like the Sloan School type of concept. Isn't it immoral to want to be famous? Maybe that's a decision you can make. But whenever I think about the question, I somehow think of the idea that your ideas are like your children. You want to be sure that they have the best life possible. So if they're not packaged well, they won't. I'm also reminded of an evening I spent at a soiree with Julia Child. Julia, and there's me. And I have no idea how come I got to sit next to Julia Child. I think they thought I was one of the rich Winstons. The Winston flowers, or the Harry Winston diamonds or something like that. There I was, sitting next to Julia Child. And the interesting thing-- by the way, did you notice I'm now telling a story? The interesting thing about this experience was that there was a constant flow of people-- happened to be all women-- people going past Ms. Child saying how wonderful she was to have made such an enormous change in their life. Must have been 10 of them. It was amazing. Just steady flow. So eventually I leaned over to her and I said, Ms. Child, is it fun to be famous? And she thought about it a second and said, you get used to it. And that had a profound effect on me, because you always say, well, what's the opposite like? Is it fun to be ignored? And the answer is, no, it's not much fun to be ignored. So yeah, it's something you can get used to, but you can never get used to having your stuff ignored, especially if it's good stuff. So that's why I commend to you this business about packaging ideas. And now you see that 6034 is not just about AI. It's about how to do good science. It's how to make yourself smarter, and how to make yourself more famous.
MIT_6034_Artificial_Intelligence_Fall_2010	23_Model_Merging_CrossModal_Coupling_Course_Summary.txt	PATRICK WINSTON: Well, don't stop. Shoot. I guess we've got to stop. I will soon go into withdrawal symptoms that will last about six weeks. But, on the other hand, we all are beginning to develop a sort of tired and desperate look. And perhaps it's a good thing to get the semester behind us and go into solstice hibernation. Anyway, there's a lot to do today. I want to wrap up a couple things, talk about what's next, and maybe get into some big issues, perhaps a little Genesis demo, that sort of thing. So here's what we're going to do first. Last time, I talked about this whole idea of structure discovery. And really, the whole reason I cracked and started talking about basic methods is because of the potential utility of taking that idea one step further and finding structure in situations where you might not otherwise find it. It's still an open question about whether that's best way to think about it. But here it goes. Imagine you've got a couple of stories. And these circles represent the events in the story. And now what you'd like to get out of these stories is some kind of finite state graph that describes the collection of stories. So you might discover, for example, that these two events are quite similar. And these two events are quite similar. So you might use that as a basis for speculating that maybe a more compact way of representing the stuff in the story would look like this. Where this one-- let's see. This one goes with this one, this one goes with this one and there's a possibility of another state in between. So that's the notion of Bayesian story merging. Now I'd like to show you a little bit more convincing demonstration of that. Here's how it goes. So there are the two stories. This is just a classroom demonstration, no big deal. But you can see there's a sort of parallel structure in there. So this is the work of a graduate student, Mark Finlayson, who processed those stories to produce those kinds of events and those kinds of events that get assembled into two story graphs. And the question is, is the most probable way of explaining that corpus of stories? And of course, the answer is no. If you merge some things like chase and stop, then you get a simpler graph, one that is more probable in the same sense that we discussed last time. Then you can merge run and flee because they're similar kinds of events. And finally, you've got think and decide. Boom. There is your story graph. And this is the same idea taken several levels higher that produce the capacity to discover, in these two stories, the concept of revenge, as promised at the beginning of our last discussion. So sometimes the Bayesian stuff is the right thing to do, especially if you don't know anything. But sometimes you do know stuff. And when you do know stuff it's possible that you can do something very much more efficient. This sort of thing takes clouds of computers to process. But we learned a lot of stuff in the course of our development that we don't use a cloud of computers to figure out. We learned how to associate the gestures of our mouth with the sounds that we make, for example. So I want to spend a minute or two talking about some work that someday will be the subject of a couple of lectures, I think. But it's the question of how to use multiple modalities and correspondences between them to sort out both of the contributing modalities. That sounds contradictory. Let me show you an example. This is the example, a zebra finch. And it's showing you the result of a program written by Michael Coen, now a professor at the University of Wisconsin. So the male zebra finch learns to sing a nice mating song from its daddy. And this is what one such zebra finch sounds like. [BIRD SONG PLAYING] Nice, don't you think? And here's what was learned by a program that uses no probabilistic stuff at all, but rather the notion of cross-modal coupling. [BIRD SONG PLAYING] Can you tell the difference? It's not known if this particular song turns on the female zebra finch. But to the untrained human ear, they sure sound a whole lot alike. So how does that work? Here's how that works. Well, I'm not going to show you how that works. What I'm going to show you is how the classroom example works, the first chapter example in Coen's Ph.D. Thesis. Here's what happens. When we talk, we produce a Fourier transform that moves along with our speech. And if we say a vowel like aah, you get a fairly constant Fourier spectrum. And you can say, well, where are the peaks in that Fourier spectrum? And how do they correspond to the appearance of my mouth when I say the vowel? So here's how that works. So here's the Fourier spectrum of a particular vowel. And when you smooth that, those peaks are called formants. And so we're just going to keep track of the first and second formant. But when I say those things, I also can form an ellipse around my mouth when I say them. And when I form an ellipse around my mouth when I say them, that gives me this second modality. So the question is, is there a way of associating the gestures that produce the sound with the sound itself? Well, there's the human data conveniently provided by a variety of sources, including Michael Coen's wife who produced the lip contour data on the right. So that's all marked up and color coded according to the particular vowels in English. I guess there are ten of them. So we humans all learn that. But guess what? We don't learn it from this. Because we don't get to work with any marked up data. We learn it from that. Somehow we're exposed to the natural world, and we dig the vowel sounds out. It's fantastic how we do that. But we do have cross modal coupling data and maybe that's got something to do with it. So here is a particular cluster of sounds. And what I want to know is, can I merge any of these two clusters to form a bigger cluster with a corresponding meaning? So what I can do is, I can say, well, I can watch these. I know what the lip form is when a particular sound is made. So I have these correspondences. So maybe there are four of those, like so. And maybe this same guy projects a couple of tones into that. And the question is, can this guy be combined with any of these guys? And the answer is yes. If they're close together on one side, maybe that suggests you ought to cluster them on the other side. But there's a question about what close means. So let's suppose that we also look at how these guys, these other two guys, project. And suppose this guy projects twice up here and once over here. And this guy down here just projects like crazy into that guy. Which are closer? These two or these two? Well, my diagram's getting a little closer. But if you paid attention when I was drawing it, you would see that this guy projects in proportion to that guy. So if we look at the-- if we take each of these projections as the components of a vector, then those two vectors are in the same direction. And the cosine between them is zero. So these are the two that are closest together from that sort of perspective of that kind of metric. And those guys are the ones who get combined. Would you like to see a demonstration? Yeah. OK, here's a demonstration based on Coen's work. So here we have two sides. We could think of one side as being vowel sounds and the other side as being lip contours or something. But you don't see anything in the diagram so far about how these things ought to be sorted out into groups. So if I just take one step, why, it discovers that those two guys had the same projection pattern as each other. So if I take another step and do the same thing on the other side, and now in the third step, the two areas that were formerly combined now form a super area. And they're seen to project in the same way as the blue area. So using this kind of projection idea, I can gradually build up an understanding of how those regions go together. And I discover, in this contrived example, that there's a vertical arrangement on the right side that corresponds to a horizontal arrangement on the left side. Now, you say to me, I'd like to see something a little bit more like the lip contour data. I'm just stepping through here until I get something I kind of like. Oh, that sounds good. That seems good. So there's a correspondence here. This is all made up data, Gaussians of various shapes and orientations. Let's see what happens when I run the clustering algorithm on that. Something definite was learned at every step. We find the correspondence between the pink region on the right and the pink region on the left. In some cases, where the regions are rather blurred together, the other side is the one that helps the system figure out how things are organized. So I cite this as an example of something I think is very important. Number one, it's possible to discover regularity without being obsessively concerned with Bayesian probability. And also, that there's very likely a whole lot of this going on in human intelligence. When we emerge and begin to examine and explore the world around us, we're presented with a lot of unlabeled data that we've got to make sense of. And I believe that this kind of cross modal coupling idea is very likely to be bound up in our understanding of that world that's presented to us. It's fast, it's direct. It doesn't take thousands of data points. It just happens. And it happens effortlessly. And if this isn't built in-- if this isn't determined to be built in, you can come back to MIT in 15 years and put me and jail. Because I think this is really the way it works. So there it is. There's a couple of things to have wrapped up. And now the next thing I want to do for the rest of our last time together in this format is talk to you about a variety of things. And I'll depart from my usual practice and move to some slides. So Dave, could we have the center screen, please? So first, a brief review of where we've been and where we've come. I think in the very first class, I talked about what artificial intelligence was. And I talked about how you could view it from either an engineering perspective or a scientific perspective. I'm on the scientific perspective side. And I think, nothing against applications, but I think we'll be able to make much more sophisticated and wonderful applications if we have not only the engineering perspective about building stuff but also the scientific perspective about understanding the stuff to begin with. So both perspectives are important. And in this case you can see that they all involve representations, methods, and architectures. Dave, I've changed my mind. Could you also give me the side screen so I can see it too? So, that's that. What's next? The business perspective, which we talked about on Thanksgiving. The important idea being that the knee-jerk expectation that the commercial value of something is in replacing people is something that is not sensible in the first instance, and demonstrated to be unlikely and untrue in the second instance. The thing that turns people on from the point of view of applications is not replacing people but making new revenue, making new capability. And that at once licenses you to not have something done exclusively by a computer but something that can be done in partnership with a person. So all the important applications of artificial intelligence involve people and computers working in tandem, with each doing what they do best-- not with replacing people. So that's that. Here's what AI does that makes AI different from the rest of the fields that attempt to contribute to an understanding of intelligence. Now, we have the benefit of having a language for procedures. We have all the metaphors that we are the custodians of in consequences of knowing about programming. We have the metaphor of garbage collection. We can talk about all sorts of things with programming metaphors that are unavailable to people in other fields that are interested in psychology. We have a way to make models because we can write programs. And when we write programs, there's no question of sweeping things under the rug. We have to work out the details. And once we've done all that, then we have opportunities to experiment that are beyond the ability to experiment in most other fields. Oh, magnificent experiments are done these days in developmental psychology and all the rest of-- all the other branches of psychology, including MRI studies that probe into your brain and see how it's consuming sugar. But it's very difficult to ablate or take away some piece of your knowledge and see how you work without it. I can take the line-drawing program that we talked about and say, how will it do if it doesn't know anything about fork junctions? And we can determine an answer. But I can't reach into Sebastian's head here with a surgical procedure and take out his knowledge of fork junctions. It just can't be done. And finally, another reason why we're different is because we can put upper bounds on how much knowledge is needed in order to perform a certain kind of task. These days, with bulldozer computing, the question most often asked is, how can you get billions of the things off the web and use them. We-- I especially-- sometimes ask the opposite question, which is how little knowledge can you have and still understand a story? That's what's interesting to me. So there's a methodological slide that talks to the question of how you do artificial intelligence in particular and, I suppose, science in general, engineering in general. There's a great tendency in this field to fall in love with particular methods. And we've had people who've devoted entire careers to neural nets, genetic algorithms, Bayesian probability. And that's mechanism envy. And a better way, in my judgment, is to say, what is the problem? Scientific method-- what's the problem? And then bring the right machinery to bear on the problem, rather than looking for things to do with a particular kind of machinery. So this is the methodology that first articulated in a forceful way, by David Marr. You want to start with the competence you're trying to understand, then bring a representation to bear on it, a representation that exposes the constraints and regularities. Because without those, you can't make models. And without those models you can't understand it, explain it, predict it, or control it. So it seems to make sense from a kind of MIT, model-centered point of view. And only when you've got all that straight do you start working on your methods and implement an experiment and then go around that loop. So that's all I want to say by way of review, I suppose. I want to take a minute or two and just remind you of what's on the final. And there's nothing to remind because you all know what's on the final already. We'll have four sections corresponding with four exams. Then we'll have a fifth and final question that will be everything else. All that stuff you slept through will be featured there, as well a little problem on Bayesian inference. We rearranged the subject, mostly so I could write the demonstrations. So that the Bayesian stuff didn't come before the fourth quiz. Therefore the Bayesian stuff that you see on those previous quizzes is likely to be harder than the stuff that we'll ask on the final, because you haven't had as much experience with it as people did last year. I've got a few icons on there to remind me to tell you a few things. As always, open everything except for computers. You can wear a costume. You can do anything you like as long as it doesn't disturb your neighbor, within reason. Well, I guess if it doesn't disturb neighbor, it is within reason. So maybe that's all I need to say. I'm not sure where we're going to be. But it's certainly the case that, historically, there are no visible clocks. So, we soon run out of all of our cellphones, wrist watches, and other time pieces as we hand them out. So it pays to remember to bring some kind of timepiece, because we won't be able to convey the time very well. And finally, I see a little calculator there. I don't recall any exam where you actually needed a calculator. But it's sort of a security blanket to have one. People sometimes see a problem and say, oh my God, what am I going to do? I left my calculator at home. So as to avoid that anxiety, you might want to bring one even though you won't need it. So that's the final. I'm sure there are no questions. Are there? It's obvious. Everybody will do well. Two shots, that whole thing. Now, what to do next? Suppose this subject has turned you on. There are a variety of things that you should be thinking about doing next semester. And I wanted to review just a few of those. One of them is Marvin Minsky's subject, Society of Mind. It's very different from this class. There are no prepared lectures. Marvin doesn't rehearse. He doesn't think about what he's going to say in advance. It's like this except it's just a conversation with Marvin. So many people find themselves bored stiff for two lectures out of three. But then in the third lecture, Marvin will say something that you'll think about for a year or for the rest of your life. That's what happens to me. I'm bored stiff two out three times. And then the third lecture he says something, and I think about it for at least a year and maybe permanently. So it's an opportunity to see one of MIT's true geniuses think out loud. So it's an experience that you don't want to miss because that's what you come here for, is to see the geniuses think out loud. Speaking of geniuses, then there's Bob Berwick. And he heroically is doing two subjects in the spring. Both of which I'd take if I could. One is his subject on Language Understanding. And the reason I'd take that is because I believe that language is at the center of any explanation of our intelligence. So that's the subject I would be, I suppose, most inclined to take if I were you. Well, maybe Minsky's. It's hard to say. And incidentally, very heroically, Bob is also teaching a course on how evolution works, how it really works-- in so far as we know how it really works-- as well. So both of those will be offered in the spring. I don't know how he does it. I don't know how he does two all at the same time. Of course there are lots of places where you can go to school, and the faculty will be teaching five courses at the same time. I just think they're crazy or something. I don't know how that works. Gerry Sussman will be teaching his Large Scale Symbolic System subject. That's sometimes-- oh, I forgot what he wasn't able to call it, something that wasn't politically correct about programming for people who really, really like to program. It's a splendid course on how to build really big systems. And we use the ideas in that subject in our research system, because it's the only way-- understanding how that works is the only way that you can build systems that are too big to be built. I may say a word about that a little later. So those are my favorite three/four picks. But there's lots of other stuff, too many things to cover-- the media lab, [INAUDIBLE] psychology. There's tons of stuff out there. And I would only mention that courses that have those three names on them are bound to be good. These are colleagues that I think I have a important perspective-- not necessarily one I agree with, but an important perspective that you should understand-- Richards, Tenenbaum, and Sinha. And now we come to my spring course, the Human Intelligence Enterprise. It's 6.XXX not because there's anything pornographic about it but because for a long time, I couldn't remember it's number. So I developed a habit of referring to it as 6.XXX and it seems to have stuck. Here's what that's about. Yeah, that might be interesting. It's taught like a humanities course, though. No lectures, I just talk. And all the TAs are veterans of that class so if you want to know if you should do it, you have several resources. You can talk to them. Or you could look at the sorts of things we talk about. Here are some things we talk about by way of packaging. Yeah, I can hear a little tittering there because people have discovered the last element. Some people take the whole subject because they want to be present for that unit. And we talk about all those kinds of things. And we look to see what the common elements are in all those kinds of packaging problems of the sort that you will face over and over again when you become an adult, no matter what you do. If you become a business person, an entrepreneur, a military officer, a scientist, or an engineer, that packaging stuff will often make the difference between whether you succeed or don't. And then, that's the second way you can figure out whether you want to take the subject. The content, the TAs, and then of course you can always appeal to the Underground Guide. And that's why it's very rare for someone that takes 6.XXX who hasn't been at this final lecture because they read the Underground Guide. Here is an element that appeared in the Underground Guide a few years back. There are no exams. But there is a tradition of hacking the Underground Guide. So this is another example of something that appeared. So it all came about because early in the teaching of 6.XXX, I was whining to the students about the fact that I've been at MIT for a long time, since I was freshman. And I still have yet to have any person I report to say anything about my teaching-- good, bad, indifferent. Nothing. Not a word. So the students decided that it would be interesting to see if they could say something sufficiently outrageous to force a conversation between the department chairman and me. And so far they've been totally unsuccessful. And I've tried everything. Winston shows up late if he shows up at all. Good instructor but constantly sipping from a brown paper bag. All kinds of stuff. But there it is. It's a lot of fun. It's a little oversubscribed so we have to have a lottery and there's about 50% chance and so on and so forth. But many of you will find it a good thing to do. Oh, yeah. And now I also want to remind myself that there is a IP event that's become kind of MIT tradition. It's the How to Speak lecture that I give. This year it'll be on January 28 in 6120. 6120 holds about 120 people and about 250 show up. So if you want to go to that lecture you should show up 15 minutes early. That's a little secret just between me and 6034 students. It's about packaging, too. It's a one-lecture version of 6.XXX. But it's very nonlinear because one thing that you pick up from that one hour may make the difference between you getting the job and some other slug getting the job. So it's one of those sorts of things that can make a big difference in a short period of time. You may sleep through 50 minutes of the 55 minutes in that lecture and stay awake for that one magical five minutes when you learn something about when to tell a joke or how to open a lecture or how to conclude one or a job talk or a sales presentation or anything. And that will make it worthwhile for you. And then of course there's the possibility of your-- anybody who's doing [? UROP ?] with me is likely to be interested in what I've recently come to call a strong story hypothesis, something that we've talked about from time to time. That's what makes us humans, and that's not what they are. They're orangutans or chimpanzees, even though the DNA that we share is-- it goes up and down. At one point it was 96%, then it went to 98%. Now I think it's back down to 97%. But whatever we are, it's not because of a huge, massive difference in DNA between us and our cousins. So in my group we build a Genesis system, modestly called. And that's what it looks like. And it has in it all the sorts of things that we've talked about from time to time in 6034. And it's about to move into areas that are especially interesting, like can you detect the onset of a possible disaster before it happens and intervene? Can you retrieve presses based on higher-level concepts? Things of that sort. Would you like to see a demonstration? OK. So you've see in a little bit of this before. In fact, I'm not even sure what exactly I've already shown you. Let me just get over there to see what goes on here. What I'm going to do right now is I'm just going to read about Macbeth. A short precis of the plot. Not the whole thing, of course. Just a few sentences about the plot. Right now what it's doing is absorbing the English and translating into a sort of internal language. A sort of universal, internal language that's all about trajectories and transitions and social relationships and all that. So it's being read there by two different persona. Each of those personas has a different educational background, you might say. They might represent different cultures. So eventually they build up graphs like that. And everything in white is stuff that's explicit in the story. And everything that's in grey is stuff that's been inferred by the system. So there are several layers of understanding. One is what's there explicitly. And the other thing that's there is stuff that's readily inferred. And because these persona have different educational backgrounds, you might say they see the killing of Macbeth at the end of the play in a different light from one another. One sees it as an act of insane violence and the other sees it as a consequence of a revenge. So once you've got that capability, you can do all sorts of things. For example, you can ask questions. So let me arrange it to ask-- by the way, this is a live demonstration. I'm thrilled to pieces that it actually works so far. Let's see. Why did Macbeth kill Duncan? You all know why, right? Yeah, you're right. He didn't actually do that. On a common sense level, neither Dr. Jeckll nor Mr. Hyde have an opinion. On a reflective level neither Dr. Jeckll more Mr. Hyde have an opinion. That's because it didn't happen. So we call these two persona Doctor Jeckll and Mr. Hyde. And you're ready to complain right away about the spelling of Jeckll, aren't you? Well, that's because with this spelling the speech generator makes it sound a little bit more like when we say Jekyll. But what really happened is that Macduff killed Macbeth. On a common sense level, it looks like Dr. Jeckll thinks Macduff kills Macbeth because Macduff is insane. It looks like Mr. Hyde thinks Macduff kills Macbeth because Macbeth angers Macduff. On a reflective level, it looks like Dr. Jeckll thinks Macduff kills Macbeth as part of an act of insane violence. It looks like Mr. Hyde thinks Macduff kills Macbeth as part of acts of mistake, Pyrrhic victory, and revenge. Isn't that cool? I bet you'd get an A if you had said that in eighth grade. But once you've got this ability to understand the story from multiple points of view, you begin to think of all kinds of wonderful things you can do. For example, you can have Dr. Jeckll negotiate with Mr. Hyde, because Dr. Jeckll will be able to understand Mr. Hyde's point of view and demonstrate to Mr. Hyde that he thinks that point of view is legitimate. Or, Dr. Jeckll can teach Mr. Hyde the subject matter of a new domain. Or Dr. Jeckll can watch what's happening in Mr. Hyde's mind and avert disaster before it happens. So let me just show you another situation here. I want to turn on the onset detector and read another little snippet. This one is about-- what should I do? We'll do the Russia and Estonia cyber war. It's reading background knowledge right now. But pretty soon, in the upper left-hand corner, as it begins to read the story, you'll see it spotting the onset of potential revenge operations or potential Pyrrhic victories, and show their foundations begin to emerge and giving the system an opportunity to intervene. So there you can see all the things that it thinks might happen. Not all of them do happen. But some of them do. And you'll note, incidentally, that this is another case of Dr. Jeckll and Mr. Hyde having different cultural perspectives. One's an ally of Russia and one's an ally of Estonia. One sees it as unwarranted revenge and the other sees it as teaching a lesson. So, I don't know. What else have we got here? Oh yeah, president recall. A long time ago, we talked about doing information retrieval based on vectors of keyword counts. That's cool but not this cool. This is doing it on vectors of concepts that appear in the stories, such as revenge, even though the word revenge doesn't appear anywhere. So because we're able to understand the story on multiple levels, we can use those higher levels that don't involve the words in the story at all to drive the retrieval process. So all that is a consequence of a variety of things, one of which is the specialists that translate the English into an internal language. And it's also, incidentally-- I mentioned it a little before-- it's also a consequence of our use of Gerry Sussman's propagator architecture. So a student comes into our group and says he wants to do something. And we say OK, here's how the system is organized. It's like a bunch of boxes that are wired together. So you get a box and we'll tell you what the inputs are going to look like. And we'll tell you what we want on the outputs. And if you don't like the inputs, just ignore them. And if we don't like your outputs, we'll just ignore that. So nobody can screw up anything because they have a very circumscribed piece of the system to work with. So I can say, for example, the president wanted Iraq to move toward democracy. And bingo. That starts a propagation through that network. All this would be unconvincing, in my view, if it weren't eventually connected to perception. Because if it's not eventually connected with perception, it's yet another system that demonstrates how smart it can seem to be without actually knowing anything. So another half of what we do is an early stage attempt to connect the language stuff with things that are going on in the world. So we say, imagine a jumping action. And there is a jumping action that's part of a test sweep developed by the Defense Research Projects Agency to drive what they call the Mind's Eye program, which was developed largely as a consequence of work done here at MIT, focused on the idea that if we're going to understand the nature of intelligence, we have to understand how language is coupled into our perceptual systems and how those perceptual systems can answer questions posed to them by the language system. That's a little demo of the Genesis system. Here are the issues that we're trying to explore. Nothing too serious, just the nature of what is extraordinarily fundamental to any explanation of human thinking. Now, all of this might turn you on. And you say to me, well, you're sick and tired of MIT. You'd like to go somewhere else for graduate school. So now that I've demonstrated what we do here, one of the many things we do here, I'll talk a little bit about other places you can go. This is a sort of MIT-centric view of the world. It represents all of the places you could go when I was a kid. But while I've got this particular diagram on here, I just-- sort of testing my MIT arrogance-- I remember a story often told by my colleague Peter Szolovits. He says that when he came to a job interview from Caltech to MIT, he was sitting here for three days and nobody spoke to him. So eventually he said, I've got to do something. He walked up to a graduate student and said, hi my name is Peter Szolovits. I'm from Caltech. And the graduate student said, Caltech sucks, and walked away. Anyway, we've populated all these places now that you see here, and more. This is just a list that I scratched up this morning. I'm sure I've forgotten many that have equal right to be on this list. But in the end, which one you go to depends on who you want to apprentice yourself to. Because a graduate school is an apprenticeship. And that means if you go to a place with just one person, it's OK if that's the person you want to apprentice yourself to. Each of these places has a different focus because they have different people. So you need to find out if there's somebody at any of these places. It doesn't matter if it's AI or some other field. Theoretical physic-- you've got to find out if there's somebody at that place you want to apprentice yourself to. So those site visits are really important. And I would like to also stress that when you make your application to graduate school, it's very different from applying to undergraduate school. Because they don't care whether their school is good for you at all. They only care about one thing-- whether you're good for their school. So don't get confused and talk about how it's a wonderful fit for you. Because what they're interested in is whether you're going to contribute to their research program. Oh, I should say that if you're applying to artificial intelligence that means you don't say, I'm interested in all aspects of thinking. You need to be focused. There's another reason why you don't say that you're interested in all aspects of thinking and that is the defect theory of AI career selection. It seems to be the case, strange though it may seem, that people in artificial intelligence often specialize their research on the things that they don't do very well themselves. So people who study language, with the exception of Bob Berwick, often have trouble getting out a coherent sentence. And people who do hand-eye coordination are the sorts who spill their coffee. So don't say you want to study all thinking because-- The most extreme case of this, though, is-- if you don't mind, I'll tell you a story about an extreme case in this. We had a visitor from Japan in the old artificial intelligence lab many years ago. He came for a year. Let's call him Yoshiaki, just to pick a name. Yoshiaki spent a year at the artificial intelligence lab, and he left his wife in Japan. And the reason was, she was pregnant. And at that time, you could not get a visa to the United States unless you had a smallpox vaccination. And because she was pregnant, she didn't want to get a smallpox vaccination because there's a small danger to the fetus if you get a smallpox vaccination while you're pregnant. So she stayed back there. So Yoshiaki, let us call him-- it was a day before he was to get on the airplane to go home. I walked into his office and his desk was covered with pictures of his wife. By the way, Yoshiaki, I should tell you, is a computer vision guy, interested in object recognition. So you might suspect he has some problem. So he's looking at these pictures. I thought, oh my God, this is a tender moment. He's anticipating his return to Japan and reunion with his wife. So I muttered something to that effect. And then he looked at me like I was the king of the fools. And he said, it's not a question tenderness. I'm afraid I won't recognize her at the Tokyo airport. So I said, Yoshiaki. How can this be? You study computer vision. You study object recognition. This is your wife. How can you think you wouldn't recognize her at the Tokyo airport? And then he looks at me, and-- God is my witness-- he says, they all look alike. Well now as we come close to the end, what are the big questions? Is it useful? Of course it's useful. It's part of the toolkit, now, of everybody who claims to be a computer scientist. What are the powerful ideas and these things? Well, here's the most powerful, powerful idea is the idea of powerful idea. And here are a few of my favorites. No surprises there. That's just Winston's picks. But there's one more I would like to add. And that is all great ideas are simple. A lot of times we at MIT confuse value with complexity. And many of the things that were the simplest in this subject are actually the most powerful. So be careful about confusing simplicity with triviality and thinking that something can't be important unless it's complicated and deeply mathematical. It's usually the intuition that's powerful, and the mathematics is the [INAUDIBLE] element. Sometimes people argue that real intelligence is possible. One of the most common arguments is, well what if we had a room? And you're in the room and you're asked to translate some Chinese documents. You've got a bunch of books. And in the end you could do the translation. But you cannot be said to understand Chinese. This is the argument of Berkeley philosopher named Sorel. So the trouble is, it's also true-- well, the argument is the books aren't intelligent. They're just ink on a page. And the person is just a computer, just a processor. It doesn't actually know anything. So since it can't be in either the person or the books, it can't be. And that just forgets that there's a magic that comes about when a running program, when a process, executes in time over knowledge that it continually contributes to. So the reductionist arguments are among the many that have been ineffectually posed to argue that artificial intelligence is impossible. But that bears longer discussion. Let me just bring up the biggest issue in my mind, which is, it's not the question of whether we humans are too smart to have our intelligence duplicated or excelled in a computer. It's a question whether we're smart enough to pull it off. I once had a pet raccoon. Now, it's illegal to have a pet raccoon. But this one was an orphan. Its mother had been hit by a car or something. A friend of mine brought the raccoon to me knowing I kind of like animals. And I have to say, I kept the raccoon for a year. At that point, she wanted to go out and be on her own. So I had this raccoon. And this raccoon is smarter than any dog I've ever had. Within a day, she learned how to pry the refrigerator door open. So I spent that whole year taping the refrigerator door shut every time. And then, we'd play jokes on each other. She wouldn't eat hot dogs. And I wanted her to eat hot dogs desperately because they're cheap and easy to serve. All she would eat was cooked chicken wings, wouldn't eat hot dogs. So one day I said, well, I'm going to play a trick on her. I took a chicken bone. I stuck it in the middle of a hot dog and put it in a garbage can. And she went for it. Her genes took over and she went for it. And she was happy with hot dogs ever after. She wouldn't let me read. She would crawl up underneath the book and interfere and make me-- she always wanted to suck on my thumb, which turned blue eventually. You'd be amazed at how much a raccoon can suck. It's just extremely powerful. The best parts were when she would go bike riding with me. I put on a heavy sweater because they have a pretty good grip. I'd put on a heavy sweater and she'd kind of mount herself on my back and look out over my shoulder, stopped traffic for miles around. So she was really smart. But the interesting thing is that at no point did I ever presume that that raccoon was smart enough to build a machine that was as smart as a raccoon. So when we think that we can, it involves a certain element of hubris that may or may not be justified. Well, there it is. Just a couple more things to do. One of which is, you should understand that Kendra and Kenny and Yuan and Martin and Gleb are doing a lot of stuff that's outside their job description. All of these quiz review deals that they've arranged are not in their job description. I didn't ask them to do it. That's all just plain old professionalism. So they've been wonderful to work with and I'd just like to-- [APPLAUSE] --offer them a round of applause. And of course, Bob and Randy and Mark have done fabulous stuff as well. And we of the staff, the TAs, Mark, Bob, and Randy, have nothing else to do except wish you good hunting on the final and a good long winter solstice hibernation period after that. And that is the end of the story. And we hope you live happily ever after.
MIT_6034_Artificial_Intelligence_Fall_2010	MegaR2_Basic_Search_Optimal_Search.txt	The following content is provided under a Creative Commons license. Your support will help MIT OpenCourseWare continue to offer high quality educational resources for free. To make a donation or view additional materials from hundreds of MIT courses, visit MIT OpenCourseWare at ocw.mit.edu. PROFESSOR: This week's problem is from the 2008 quiz one on search. It is motivated by the search of evil overlord, Mark Vader, who is shopping for a new evil stronghold. He starts from his current stronghold, which is S, the Depth-First Search Star. Now the Depth-First Search Star has the following qualities. It has a small, thermal Exhaust Pipe Weakness. It has the quality of That's No Moon, but it does not have a Race of Enslaved Minions, or a Secret Escape Route, or Sharks with Laser Beams. Now when I originally wrote this problem in 2008, there was another quality that very important. It was a giant Laser, and it did have that property, as did Dr. Evil's Moon Base, but that was removed to make it easier to solve. So we're going to solve it without the giant laser, that we can just have the Sharks with Laser Beams. So Mark Vader has gone to Ackbar's Emporium of New Evil Strongholds, which are listed on the left, and he's trying to figure out the best way to get from his current stronghold to his goal stronghold, the 603 Fortress, which has the admirable qualities of no small thermal Exhaust Pipe Weakness, and it still has That's No Moon. It has a Race of Enslaved Minions, and a Secret Escape Route, and Sharks with Laser Beams. So it's got everything you would want and no weakness. However, he can only move between fortresses that have exactly one difference. Fortunately, Mark remembers how to perform the search techniques he learned in 6.034 from his mentor, Emperor Palpatine. So we've got several fortresses here. We've got the Depth-First Search Star. We've also got Shaoul Ghoul, which has the qualities, basically, of That's No Moon, and none of the other qualities. Dol Guldur here, has got the Exhaust Pipe Weakness somehow, despite being Sauron's stronghold. It's not a moon, and it has those enslaved minions. I guess the Orcs are enslaved. Moonraker, here, only has the Exhaust Pipe Weakness, so it's not very good. But James Bond didn't need to deal with stuff. Zeal Underwater Palace has That's No Moon and a Secret Escape Route. Zeromus' Lunar Core has an Exhaust Pipe Weakness and a Race of Enslaved Minions. Whalers of the Moon Ride has the Exhaust Pipe Weakness, Race of Enslaved Minions, and Sharks with Laser Beams. 6.03 Fortress we've already been over. Atlantis has all qualities except for a Secret Escape Route. Willy Wonka's Factory has some of everything. It is, after all, Willy Wonka's Factory. Highlight, the Race of Enslaved Minions, the Oompa Loompas. And Dr. Evil's Moon Base is only missing That's No Moon, because it is a moon, because it's a moon base. So if you got all the references that were used in making those fortresses, I'm sorry. There's nothing I can do for you. Now, Mark, being a clever evil overlord, realizes that he can produce a graph of the exploration choices with edges joining strongholds that differ by just one feature. So although this is a graph here, as Patrick often says-- and you get to hear again right now-- search is about choice, not just about maps. So we're not moving around anywhere in the real world, but we certainly are going to move around this graph to pick between these many stronghold choices. How can we decide where to travel? Well, we're about to find out. So we've got this lovely graph right here, and now we're going to get to do depth-first search on the graph. Now there's a lot of ways to do search, and you've seen Patrick do it, you may have seen some people do it in recitation. And I'm going to offer you guys a unique opportunity to see me do search in one of several ways, or possibly more. I have the tried and true method. You have the queue, or agenda, or whatever you want to call it. You keep track of it at every level. You make damn sure that everything you're doing is right. It takes a long time, and you get the answer right. Or there's one where you only work with the goal tree, and you draw it really fast, and you might make a mistake, but it's going to solve it more quickly. So who would rather see the reliable, but slower, approach? Who would rather see the faster, but less reliable, approach? All right. Monte Carlo people here, rather than Las Vegas. If you don't know what that is, you'll learn it in algorithms class. All right. Then there's a third approach that I probably won't show you unless there's overwhelming favor for it. That is the approach that is manically fast and will solve depth-first search in no time at all, but it is very likely to make mistakes. That is the approach that I use when solving depth-first search. I've used it enough times that I don't make mistakes that often, but even I still do. So does anyone want to see that one? OK. Not that many people. Well, it doesn't take much time, but I'm just afraid that if I show it you, that you'll be like, ooh, that's the only way to do it. So we'll save that for later if we have more time. It essentially, for people that are interested, involves just using your chalk or finger to trace through the tree really fast and figure out what it's doing by drawing little tiny lines. It is very fast, but it's not very accurate. And you have shown, basically, no work if you get it wrong, so shame on you. OK. Well, let's do the somewhat faster way where we don't draw out the entire agenda, and then people who were the almost nobody who raised their hand for that way can check it out in tutorial the way that you do it using the agenda. So let's do it using the goal tree. So it's a bit faster that way. So we're going to start, starting at S, the start node, and going to G, the goal node. That's a standard notation, but make sure that when you're taking the test, you check to make sure where the start node is, where the goal node is. It's pretty much an asshole move to have a G node that isn't the goal node, but sometimes there won't be an S node. That's a pretty good clue that something else is the start node. Now there's a few little white star ideas that we have, or silver star ideas, for search. One is lexicography. What it lexicography? Who cares? Well, the most famous lexicographer I know of is Samuel Johnson, who wrote a really famous dictionary. And the only reason we care about it in this class is because you'll always see the instructions "break ties in lexicographic order." And then you might be like, wow, that's really rather a wordy word, or a sesquipedalion word, or whatever really long word you want to use to describe the word "lexicographic." What the heck does it mean? Basically, it means alphabetical order, like you would do in a dictionary. So for instance, in alphabetical order, A comes before B. So you would go to A before B. No, even beyond that point, there's several ways to do a lexicographic tiebreak. And it's not always consistent between algorithms how you would do a lexicographic tiebreaker. In the most recent time that Patrick was kind of randomly talking to the staff, he made it clear that the way he would like us to do it, and therefore, the way we'd like you to do it, is to tiebreak based on the very latest nodes at the end. So you might say, that's not really lexicographic. Wouldn't SAB come before SGA in the dictionary? And my answer to you is, yes, it would. Yes, yes, it definitely would. However, that is apparently what we're doing, according to the last thing that Patrick said, and we will keep you updated if he decides to go back to dictionary order. So that's lexicography. You might be wondering what the picture here is. Well, this is our friend, the ouroboros. Long has it been a symbol in alchemy of infinity, eternity, or maybe a sort of infinite energy engine, turning iron into gold. But today, unfortunately, the ouroboros is an endangered species because no biting your own tail in 6.034. A lot of people mess up from this. It's an honest mistake. I mean, we've gotten some emails from staff members who couldn't solve one of the problems because they forgot about this. So it's OK if you forget, but try your best to remember. No biting your own tail is the only smart thing our system does. You'll remember last week when we were talking about rules, and we were saying that the rule system is not a smart creature. It doesn't know that "not Polly is dead" should be the same as "Polly is not dead", or something like that. It's too dumb to figure that out. Well, this system is also dumb, but one thing it knows is that Patrick really hates if the same node appears twice within the same path. And it will destroy that immediately, before even adding it to the queue. It's gone. It's not considered. It's out of there. So very important, sad ouroboros, no biting your own tail. All right. So now that I've gotten through that, let's actually solve the problem with some depth-first search. How will we do that? Well, we're not going to use the queue, as popular demand, we're going to use a goal tree. So we'll start at node S. What are our choices at node S? Well, I'm going to force you guys to help. So it won't be quite as fast, but it'll be fun. So what are our choices at node S? You. AUDIENCE: Me? PROFESSOR: Yes. AUDIENCE: A or B? PROFESSOR: Not quite, and this is something I like about this problem. You got most of them. Do you see that there might be another choice? Everyone? Yeah, C. This is a big problem that has happened on a few different quiz problems where there's a sort of a grid that looks like a tree, or a graph that looks like a tree, where people aren't as willing to go up as they are to go down. You can see it on some of the other past quiz problems too. Make sure that you check the connectivity, and also note that unless otherwise specified-- and I don't think we've really done this much-- the little connecting edges in our graphs are bidirectional. You'll see a big arrow, and probably instructions written at least eight font sizes higher than all the rest of them, and bold, if we're going to do something different for this quiz. It happens that they do it sometimes, where I'll just say in recitation, we never do this. And then on the quiz, they do it. But it will be in giant bold letters, which also leads me to another silver star idea that I forgot to tell you guys last week. So I hope people didn't ditch out on me, like, oh, Mark's boring. We won't come next week. And then miss this one because it just came to my mind. That is, read the instructions. It is a very important thing for the quizzes, and Patrick will, in a later lecture, tell you guys, ask why five times, and he'll explain why that's a gold star idea. My parallel to this is read the instructions five times. Maybe not five, maybe three or four, but at least three. Read the instructions. Read them again. Then read them a third time only paying attention to whatever is bold and eight sizes larger than all the other ones, cause there's going to be something there that's that size. And it going to be the thing that everyone misses. So make sure you're not everyone, and you read the instructions a lot of times. But anyway, so yeah, we've got it right. S goes to A, B, and C. OK. We're looking at A, B, and C as possibilities. Everyone, help me make the lexicographic tiebreak. Where do we go? AUDIENCE: A. PROFESSOR: A. That's right. Actually, I'll call on you guys for other ones, but that first step is pretty simple. We could have everyone do it together. Everyone, A only leads to? AUDIENCE: D. PROFESSOR: That's right. Why doesn't it lead to S as well? AUDIENCE: [CHATTER] PROFESSOR: No biting your own tail. Everyone's right. Good job, everyone. All right. S leads to these three. You go to A, go to D. Dead end. Backtrack. OK. When we backtrack, we backtrack up to A. There's no other children. We backtrack up to S. Where will we go now? AUDIENCE: B. PROFESSOR: B. It's the next one alphabetically. All right. B can't go to S, so it only goes to? AUDIENCE: H. PROFESSOR: Yes. When we're at H, we can go to? AUDIENCE: [CHATTER] PROFESSOR: F or I. That's right. You guys get this. Depth-first search is easy. All right. But we choose? F. That's right. When we're at F, we can go to? E and J. That's right. We'll choose? E. When we're at E, we can go to? We can go to C. It's not on this particular path. People are correct when they say C. When we go to C, we can go to nowhere. We're dead. Backtrack. We backtrack. We can't do anything at E. We go to F. Now I forgot to tell you guys, but this is an important note. Someone's going to get this wrong. It's going to be one of you. Look around you, all through the room. At least one of you is going to do this. So, now I'm going to tell you not to do this, and then hopefully, that will still be true, but for fewer people. When I went to D, and then I backtracked, and I went to S, how many times did I backtrack? Once. I backtracked once. This would have been more obvious if we were doing the really slow boring agenda way, or queue way, because when we got to SAD, and we expanded it, and there was nothing left, we throw it off the top of the queue, and go to the next thing on the queue. And it turns out the next thing on the queue is SB. That step was only taken once. Since we're using the goal tree, which is faster, it looked like we did two, but we didn't. And if you use the queue, you'll see that because it'll go S, and then we expand that, SASBSC. SBIC OK. We expand SA. SADSBSC. We expand SAD. It's dead. SBSC. One backtrack. We're back at SB. So, so far we've done two, not four. It's pretty intuitive to say that you did two there. It turns out you didn't because of the algorithm that's backing what lets us do this goal tree search. So try to make sure you're not the one who says two. And I, on the other hand, will try to make sure that we don't take off too many points if you do. Question? AUDIENCE: So any time you backtrack, regardless of how much it chains, any backtracking is just considered once. PROFESSOR: It is always going to be considered one step. It's possible to backtrack two times in a row. Like, if B didn't go to anything, instead of going to H, then we might backtrack twice before we got to SC. Generally, anytime you draw a swizzle, if you're like me, and you draw a swizzle-- I suggest the swizzles. They're very nice little things. But any time you draw a swizzle on your graph, you have backtracked once. You could even write BT next to the swizzle, and go back and count those, or even just count the swizzles at the end. You've got it. You got the answer. Because they often ask, how many times have you backtracked? All right. So good questions, everyone. So anyway, we backtracked from C. Nothing at E. We go back. We were at F. So we were at SBHF. We go to J. It's the only choice. At J, we can only go to? AUDIENCE: I. PROFESSOR: And at I, we can only go to? And we win. It's not an optimal search, so as soon as we see anything with a G on the queue, boom. Winner. All right. And we're done. We did it. That wasn't too bad. For those of you who are vaguely interested in seeing how I would do it the really super fast way, it goes something like this. All right. ABC. A comes first. Only D. Backtrack. All right. B comes first. H. F comes first. E. C. Nothing. Backtrack. We came that way. JIG. So that's the really fast way to do depth-first search. Don't do that, kids. We don't like-- well, maybe you can. If you get it right, we're not going to take off points usually. Generally, when we say draw the goal tree below, it's just assigned partial credit. However, we are pretty strict about that. If anything is wrong, except for maybe, like, OK, you write everything exactly right, and forget G at the end. If anything that's not completely understandable is wrong, you will probably lose all of the points, and it'll be a lot of points if you don't draw the goal tree. I will emphasize drawing the goal tree is a good idea. All right. So now let's do a breadth-first search. Before we do the breadth-first search on this tree, I will tell you guys that there is also a fast way to do the breadth-first search which is less risky, and it really depends on how nice they are about what they ask for. In this case, the breadth-first search question asks, what path does Mark find using breadth-first search? Rather than saying, what nodes does he expand in order, or anything like that. That is important. If that question is asked you, there is a trick that will let you solve it very, very quickly. In fact, faster than depth-first search. You can solve it by inspection in about 30 seconds. Does anyone know what the answer is? What path did he find? AUDIENCE: [CHATTER] PROFESSOR: That's correct. So the answer is? On this graph, the answer is? AUDIENCE: SBHIG. PROFESSOR: SBHIG. That's the answer. You would have your five points. Did people see that? It doesn't always work, so we're going to actually solve it. But did everyone see that sometimes you can get away with not doing it? Because breadth-first search is guaranteed to give you the path, as we heard correctly, with the least number of jumps, and if there are more than one that tie with the least number of jumps, you can just lexicographically figure it out, in this case with actual dictionary order. But SBHIG is the only one. Let's do an actual breadth-first search though, so we can feel better about ourselves. OK. So you've got S, and S goes to, as we saw before, ABC. Maybe that doesn't have to go quite so high. All right. As we saw before, S goes to ABC. And you already told me that A goes to D, and you told me that B goes to H. But what does C go to? E. That's right. As you can see, we're expanding it level by level, left to right. All right. So SAD, were does SAD go to? SAD goes to nowhere. It's dead. SBH. Now, wait. You might say, wait a minute. Patrick said that we're using this weird dictionary order where E is at the end that comes before H. That's our tiebreak order, but it turns out that breadth-first search and depth-first search don't sort in any way. It's very important. They don't sort the paths that are currently on the queue. So you're going to just go left to right, left to right, left to right. And only at each node are you going to break ties in lexicographic order. All right? So SBH. H goes to, as we already know, F and I. E, we think we already know, but we don't quite because this is E coming from the other direction. E, this time goes to F. That's right. Well, actually, you guys do already know it. All right. Great. Now we come over here. HF goes to, as we already know, E and J. I goes to, as we don't already know, HI goes to G and J. That's right. And as it turns out, by an implementation detail, we're done. Questions? Does it not expand depth at all? Now this is an implementation detail. It's perfectly sane and reasonable to make a breadth-first search that likes to finish its entire level that it's working on. However, in our implementation, and we would have seen this if we had been pedantic and drawn out the entire queue-- that's another reason why drawing out the entire queue is, as I said, more reliable. In our implementation, since it's not an optimal search, the moment anywhere on the queue, you add something with a G at the end, you finish. And because of the fact that the way breadth-first search does its mojo, is that it adds everything to the end of the queue. That's how it does it level by level, right? It adds it to the end of the queue, instead of the front. Well, then you'll add it to the end of the queue, and then you will have it on the queue with a G. So you won't have to do SCEF. Another question? AUDIENCE: I'm just wondering, so the breadth-first search, there was no backtracking [INAUDIBLE] had no other-- PROFESSOR: Ah. That's a good question. So the question was, so for the breadth-first search, there was no backtracking. D died. Why didn't we backtrack or something? The answer to that one is, for breadth-first search, backtracking doesn't really, it isn't really a thing, like it is for depth-first search. Why? Well, because breadth-first search, we're sending our infinite monkeys down every path. In depth-first search, we're really focused in now. We're like, we want to get there. We want to get there. Ooh, this way, this way, this way, this way. And might have gone the wrong way, and then we hit a dead end, we're like, oh, crap. And we go backwards. But for breadth-first search, we really are like an evil overlord. And we're like Mark Vader saying, storm troopers, go every direction. And then from there, go every other direction. And so, even though, yes, when we got to D, some of the storm troopers hit a dead end, and probably it was reflective, and they shot it, and hit themselves or something like that. Meanwhile, the storm troopers we sent to B and C are still fine, so we don't need to backtrack. AUDIENCE: Cause there are, like, other troops going down the tree. PROFESSOR: Yeah, there are other troops going down the tree in every direction. Whereas in the depth-first search, we only sent-- we were like, there are definitely no droids on this escape pod. Send everyone this straight direction, and we only sent them to A and D. And then, so we had to backtrack because we hit a dead end. Does that make sense? Another question? AUDIENCE: [INAUDIBLE] PROFESSOR: So the question is, in this breadth-first search, did lexicographic order ever come into play? The answer is yes, in a very subtle and sneaky way. Which is, I wrote E before J, F before I, and G before J. If I was not using lexicographic order, it might have been reasonable to write, for instance, when I was expanding I, to write J before G because J was higher up on the tree or something like that. But the only way it came into play is that I wrote them left to right in alphabetical order every time, and I wrote ABC. Another question? AUDIENCE: We see the F node twice. Suppose that the goal node wasn't reached [INAUDIBLE], then you would have visited F again. PROFESSOR: And we would have visited F again. That's correct. The question is, F is listed twice. So if we hadn't reached the goal node, let's say that down there, after I is Z, and after Z is the goal node, then would we have visited F again? The answer is yes. As I said, this is the approach that throws your troops every possible way. So there's storm troopers going from E to F, and there's storm troopers going from H to F, and they're going everywhere. Now there is a way to cut down on this. You could do breadth-first search with an extended list. If you did do breadth-first search with an extended list, that would be sort of equivalent to as soon as you expand one node, as soon you send storm troopers out of I to look at G and J, one of them stays at I, and if any other storm troopers come to I, and they're like, we want to see what's past I, he's like, no, no, no. We've already sent troops past I. We've got this. Go back home. It's OK. There are no rebels here, but it might be Han Solo dressed up as a storm trooper, as we'll see in the next problem. Where our extended list screws us over. But for this problem, do you see what I mean? With an extended list, we could avoid this because the extended list basically says, once I've expanded and searched past here, don't do it again. But if you don't have one, yeah, you'll do F twice. In fact, didn't we do something twice here? Oh, it turns out we didn't. But we easily could have. We almost did E twice. We almost did F twice. It turns out we didn't, but we could have. Another question? AUDIENCE: [INAUDIBLE] for the implementation of the queue that when you expand I, that both G and J were added simultaneously and in lexicographic order? Or is there a piece where it would be [INAUDIBLE] order, G got added before J, so hence, it stopped [INAUDIBLE]? PROFESSOR: Ha. That is a very good question that provokes a minor change. The question is, when you expand SBHI, is SBHIG added before SBHIJ because of lexicographic ordering, or are they both added at the same time? The answer is, simultaneous. When you do the expansion, you instantly receive say, a list maybe, or whatever data structure you like, containing all of the children, and they get appended to the queue at once in order. So you will receive all of them. I suppose you could create some kind of search problem where you have, like, an adversary node, like a wumpus, or you're hunting a wumpus. And if you don't want to add the exact wumpus to your search tree, or you lose. You only want to add the one next to the wumpus. I don't know. I'm making this up. And unfortunately, you would have to add all the children at once. Our current algorithm does not add one at a time. All right. So we've done a breadth-first search. Great. Now we're going to do some optimal search. That is going to require me to draw a slightly different graph. I will draw it on what is left of the bottom here. And then we'll solve it on what is left of the bottom here. All right. Mark has his new stronghold, and he wants to invade parallel universes. Now he's programmed his evil supercomputer to find the shortest path of jumps from his starting universe to the goal universe, G. It takes a certain amount of energy to move from universe to universe based on the differing factors between those universes. Like maybe in one universe, a butterfly didn't flap its wings in China, whereas in the other universe, sentient reptiles rule the earth. And so, Mark's supercomputer has tried to create a heuristic value that determines how different the universes are to guess, sort of, how much energy it's going to take to get from his start universe, S, to his goal universe, G, given that currently his armies are at a particular universe. In other words, OK, it's not A you put in search, but dammit, we're going to have a heuristic distance to the goal. And we're going to have a graph with distances. So let's see what it looks like. All right. And down here is G. So this otherwise unassuming heart shape hides an evil invasion force. So let's see. The distance here is 100. The distance here is 3, 4. The distance here is 4. The distance here is 50, 50, 14, 4, 16, 16. Oh sorry, these are connected. 30 and 10. The heuristic values are 0 at the start node, 50 at B, 60 at A, 55 at C, 50 at D, 56 at E, 50 at F, 0 at G because it's the goal node, 39 at H. I'll draw the little H smaller. And 0 at I. Well, after all, I goes right to G. It can be 0. It'll be great. OK. Now, this time Mark is trying to conserve the energy of his parallel universe jump. That, understandably, takes a lot of energy. So he needs to program in the shortest number of universe jumps that will get him to the goal. He doesn't mean the least number of jumps. He means the least amount of energy on these edges. He's not interested in getting to this new world, and not having enough energy left to blast the crap out of it. So we need to find the shortest path. So first Mark programs a simple branch and bound search. He adds in an extended list to his branch and bound just to make it a little bit faster. As usual, he breaks ties of equal length in lexicographic order. So let's list the nodes that Mark's computer adds to the extended list in order. Distances are shown next to the edges. Ignore the number in parentheses for now. They are heuristics, and we're not using them. So that brings me to another point that I'd like to drive home before we go for the home stretch and solve these problems. What the heck is the difference between branch and bound and A? OK. I like to liken it to the following. I probably should have, I'll make this a silver star idea. OK. The silver star idea is pizza. It doesn't really look like pizza, but that's the silver star idea. Branch and bound is a cheese pizza. It's simple. If you order it for a large group event of college students, they will eat it, and things will be OK. Now A search is some kind of meat lover's or supreme pizza. Maybe a meat lover's pizza. It's got all these extra toppings. A lot of people are really going to like it. It might be better. But then you've got a vegetarian, and everything's screwed up. So basically, A* is just branch and bound with some extra toppings added on. In this case, one of the toppings is an extended list, an extended list which we'll see keeps track of where we've passed through and already expanded out, and it never goes back. The other topping that we're going to add is a heuristic. A heuristic tells us about how far we think we have left until we're at the end, so we don't go through really short paths that go completely in the wrong direction. All right? Between the two of those, we get our supreme pizza, but sometimes, as we'll see, they sort of mess each other up. So one of the toppings just doesn't go well with the other one. Maybe someone doesn't Hawaiian. They think that the ham doesn't go well with the pineapple. So let's do a branch and bound that just has an extended list. Maybe we've got some green peppers on this pizza. This is going to be safe. All right. So we're going to list the nodes as to the extended list, and the way we're going to do that is, well, you guys said you like goal tree more than queue, so let's do it with a goal tree. So we've got S, and as we already know, S is the only path. Our current length at S is 0. That's the lowest of all of our lengths on the tree cause it's the only length. So we go to A, B, and C. It may be hard to see, so the length of SA is 100. The length of SB is-- AUDIENCE: You mean ABEF? AUDIENCE: Yeah. [CHATTER] PROFESSOR: Oh. You're right. It's a different tree. I was doing A, B, and C from the other tree. Thank you, friends, for correcting my foolishness. The length of SB is 3. It is a different tree. I was trying to save work that wound up creating more because someone's going to be confused by this. So the length of A, B, E, and F are 100, 3, 14, and 14. Which one do we choose? Lexicographically, it's A. We choose that one, right? AUDIENCE: [CHATTER] PROFESSOR: F is going to be 14. That' a 1. Sorry. AUDIENCE: 4. 4. PROFESSOR: Oh, it's 4? I wrote this problem. I should know it. You're right. F is 4. My apologies once more. That's what I get. All right. I'm going to stand over here from now on when I write over there. OK. So which one of these are we going to choose? Even with the 4, we're going to choose B. That's right. Lexicographic be damned. It's only for tiebreaks. We want the shortest path. Our special thing with branch and bound is we take the currently shortest path, whatever it is. Great. So we expand B. Fortunately, and I'm pretty sure I've got this correct this time-- oh wait, I've got an even better idea. I'll just take this little hand sheet with me, and then I don't have to look at that at all. OK. So once we expand B, B goes to D. Our path length from B to D is 4, so what will we write next to little G here? 7. That's right. We add them all together. I'm going to ask you guys again the next time we do this with the heuristics, which is going to be the next part of the problem, and someone's going to give me the wrong answer next time. So stay tuned for that. All right. So we've got SBD. We've got all of these. Where do we go next? F. That's right, because currently 4 is the shortest because you guys corrected me correctly. So SF. 4 is the shortest. F only goes to H. And what do we write next to the H? 20. That's right. 16 plus 4, 20. All right. Where's our current shortest? It's the D, SBD. All right? So SBD. D only goes to I, and we've got a 57. Wait a minute. I want to get this problem right, so we better actually write the extended list, [INAUDIBLE] because that's the only thing they're asking us for. So first we extended S. Then we extended B. Then we extended F. Then we extended D. All right. So far, so good. And I have the answer key, so I know we're doing it right. OK. What are we doing next? E. That's right. So we extend E. E goes to H. And when E goes to H, we've got a length of 30. All right. Who's our winner now? Which H? SFH. That's right. SFH is length 20. Oh yeah, of course I should write an E in here. I always forget to do that. SFH is length 20, so we will extend H, and I'm going to write it here preemptively. When we extend H over here, H only goes to I, with length 50. Great. What's the next shortest? The next shortest is the other H. However, will we expand it? You guessed that because I asked that question in that way. You knew the answer was no. Why don't we expand it? It's already on the extended list. That's right. So since it's already on the extended list, this one dies a horrible death. I like writing an X through it instead of writing a swizzle at the bottom. You can do whatever you want. All right, it's gone. It's not a choice. What's the next best one? SFHI. That's right. That goes to G, and the length is 60. And we've extended I. All right. Question is, are we done? The people who say no, I like you. You're smart. You realize that just because the G is on there, that we can't end. However, the people who said yes, you are either oblivious or really, really smart, and I'm going to choose to assume you're all really, really smart because the really, really smart people said, OK. Yes, we're not quite done just because we added a G. We still have to check to make sure there are no lengths with shorter path, but look. The only one with a shorter path is I, which is already on the extended list. So actually, we are done. Double check. We've got it. So these are the paths we extended in order, and our final path is? Everyone. AUDIENCE: [CHATTER] PROFESSOR: That's right. SFHIG. I claim that is the correct path. However, Mark is frustrated by branch and bound's speed. I don't know. I wasn't that frustrated. Seemed pretty good. But Mark is frustrated by branch and bound's speed, so he reprograms his computer to use A. Mark counts the number of subspace anomalies between each universe and the goal, and uses this count as the heuristic for A. These numbers are in parentheses. Hopefully, you can read them. Yes? Oh, we've got a question. Right here? AUDIENCE: So, given the implementation, you said that you expand all possible nodes. So why doesn't I go to C and D, as well as [INAUDIBLE]? Like, why doesn't it expand to [INAUDIBLE]? PROFESSOR: Ah. That is a very good question. And the answer is, a very simple answer. You guys tricked me. No, but I should have been able to figure it out. It does go to C and D. The correct tree, which we wouldn't have lost points for having the incorrect tree here cause we did get the correct answer, yes. The reason why it is, that same reason that very first time I asked someone something, he didn't remember the C. It's easy to forget to go up the tree. It does actually have a C and D. However, they are horrendous paths. They are 100 and 100 on their path length, and so it doesn't matter. But you were correct. The official answer we had up there is wrong. AUDIENCE: Would the D get added to this list of children that's already in the extended list? PROFESSOR: Good question. The question is, would the D get added to the children? After all, it's already in the extended list. The answer is, we search for, and remove, and kill all of the attempts to extend something that's already on the expanded list at the time we try to expand it. The time we try to expand it is only when it's on the front of the queue because it's the currently shortest path. So that means they get added. It's just that when it comes time to expand it, it will get crossed off no matter what. Turns out, it escaped execution because of the fact that we never expanded it. AUDIENCE: So H should go to E as well? PROFESSOR: So H should go to E as well, is the question. The answer is yes. H should go to E as well. A lot slipped past me this time. H should go to E as well, with a length of 36, and it dies there. This one will actually be checked, so it actually does make a difference. If we ask how many times was a node executed due to already being on the extended list? Very good. Very good notice. It should be on there. We'll get it right next time. All right, everyone. So I'm working together with you. I made the mistake too. Easy one to make. It can mess you up. It didn't this time. We're going to get it-- question? AUDIENCE: If G's the goal at the end, and you get G in your outcomes, and you know the number that's shorter than everything else, do you have to actually extend at G? PROFESSOR: So, the question is, do you actually extend G? Should we even put G in the extended list? The answer is, the answer to that question is, it is a matter of taste. And in the questions where we, in my opinion, rather foolishly, asked how many nodes were extended, rather than ask you to write them out, we generally accept the answer where you didn't extend or where you did extend G. It's an implementation detail. You can either have a fail safe that, as soon as it sees G at the beginning of the list, says, we're out. We're not going to extend. We're done. We win. Or a fail safe that, when you're about to extend something, and you go into the extension process, and it sees that it ends in G, that it wins. So it is a matter of taste. If you kind of like to watch your little guy doing the search win, and have an S and a G, you can put it on. If you don't, you can not put it on. We won't take off points for whether or not it has a G at the end because clearly, if you did all the other crazy stuff correct, well, you could have written in a G if you wanted to. I think, unless there's someone who mis-solves the problem right at the end, is like, oh, who cares about G? Oh no, it's stuck. It's a dead end. We lose. But that would be probably pretty rare. So let's solve the A. So I'm not going to make the same mistake twice. We go from S to ABEF. And S, by the way, had a value of 0. All right. What is the value at A? Well, I think it's 160. The question is, how do we calculate this? Well, it's the path that we've traveled so far plus the heuristic value at the final node. This is why someone's going to give me the wrong answer at BD, but let's see. So we have 160 here. OK. So SB. What is the heuristic value here? We've got 3 for the path, and 50 for the heuristic, so it's 53. SE. What have we got here? We've got 14 for the path, 56 for the heuristic, so it's 70. All right. SF. We've got 4 for the path, 50 for the heuristic, 54. OK. Who's our winner? It's B again. Barely, but it is the winner. Extended list up here. SB. All right. B, as we saw, goes only to D. What is the value that I should write here? AUDIENCE: [CHATTER] PROFESSOR: OK. I am happy I heard all the things that I expected to hear. I heard the correct answer, which is 57. I also heard someone say 107. So why is it 57, and not 107? Someone does it this way every time. Do not add up all the heuristics along the way. I will try to explain to you why you would never want to do that. The heuristic value at any given node says, given that I'm here, how much work do I think I have left to get to the end? All right? It's sort of like, let's guess the last few nodes in the path that we haven't done out yet. So you can see why it would be bad to add that for every node in your list because then you're double counting all of the last nodes. So add the path so far to the very final heuristic. That's 3 plus 4 plus the 50 that's with D is 57. So our current winner, then, is 54, with F. Just the same as last time. F goes to H. And what's our total value at H? 20 plus, that's a 39, 59. So who wins now? D, with 57. But we extended F, yes. All right. So D was 57. All right. So D was not one of the ones that I tricked myself with. D only goes to I. What's our value at I? I think our value is 57 because I has a heuristic of 0. Yes. I has a heuristic of 0. Our value at I is 57. OK? And I extended D. Who's the winner now? I? OK. So I is the winner. With I as the winner, we extend I. I goes to C, D. Ha ha. Good call, whoever back there figured out my secret error before. C, D, and G. AUDIENCE: [CHATTER] PROFESSOR: C, H, and G? Oh, you're right. We already went to D. Aha. C, H, and G. You're correct. Everyone see that? C, H, and G. Absolutely right. I goes to C, H, and G. So the path to C, H, and G is a hard path. To C, we got, what did we have? 57 plus 50 is 127. Oh, and by the way, C, G, and H. We've got to do it in lexicographic order. So 127 to C. To G, we have 67. And to H, we have 87. So who wins now? H. H, with 59. That's right. H, with 59. And as we already know, cause it was the winner last time, H, with 59, goes to I, which has, I believe, 50. Well, it's still gets added. Remember? We're only going to kill it when it gets extended. So who's the shortest? So you were a step ahead, whoever asked me that question. Who's the shortest? H also goes to E. Absolutely right. H also goes to E. I was getting ahead of myself here. H also goes to E with a length of extra 16. So we've got 20 plus 16 is 36 plus 56 is 92. All right. That's good. So who's the shortest? I is the shortest. Do we extend it? No. It's on the extended list. All right. Who's the shortest now? It's hard to read, but the shortest is this 67 on the G. So we're done. We win. Our path is SBDIG. Yeah, for A. It gets the right answer, right? No. Unfortunately not. So what happened here? Why did we not get the right answer? Be as specific as possible. What are people saying about heuristics? AUDIENCE: [CHATTER] PROFESSOR: All right. People are saying the heuristic must be consistent. That is both correct and specific, so I applaud you. Too easy to say the heuristic was inadmissible. It actually was completely admissible everywhere. So that leads us into our very last point for the day. What the hell are these heuristic consistency and admissibility things, and why do I care? Well, the reason why you care is many fold, one of which is that it's almost guaranteed to be on the quiz. But what are they? Admissibility is a check at every point to make sure your heuristic at that point, which is supposed to be an estimate of how much work you have left to do, is always an underestimate or an accurate estimate. It can never be an overestimate. Why not? Well, as Patrick showed you with his mostly correct example in lecture, if it's an overestimate, it's never going to expand that node cause it's going to think, you know, if you write one million as a heuristic, it's going to think it needs to do one million after it's done going that way. It's not going to want to go that way. All right? So it's always got to be an underestimate or an equal estimate. But let's say that you, at the quiz, and you forget Mark told me this. You didn't bring any notes. You're just having a brain freeze. You're like, oh no. Is it supposed to be an overestimate or an underestimate? I can't remember which one. How can I figure it out? How can I figure it out? I propose you the following calm, soothing mantra slash sutra that will help protect you. Think to yourself of the following question. We know that A* sometimes messes up with the heuristic. Does branch and bound always get it right? Yes. What heuristic values does branch and bound add everywhere? 0. Right? It has no heuristic. It essentially adds 0. Is 0 an overestimate or an underestimate? An underestimate. Therefore, since branch and bound always works, the one you're looking for is an underestimate. And I know from when I was taking the class, I always had a moment where I had to spend, like, two minutes convincing myself, OK, I'm going for underestimate, not overestimate. This will help. It will take fewer than two minutes to do that. So what is consistency? Consistency is a little bit stronger of a claim. When you claim that a graph is consistent, what you're saying is, between any two nodes, or to do it more simply, between any two adjacent nodes, the distance between the heuristics is less than the distance between the nodes. In other words, admissibility is a sort of, like, consistency between every node in G. Whereas consistency is consistency between every node and every other node. All right? Consistency is a stronger claim. Any graph that's consistent is always admissible. Any graph that is inadmissible is always inconsistent. That's the contrapositive. Question? AUDIENCE: Less than or less than or equal to? PROFESSOR: Equal to is still OK. Equal to is always OK. That's great. It's a perfect estimate. But never greater. Now, if a graph is inconsistent, will you lose? Will you lose? Why is it called admissible, then, if it's sometimes not admissible? Why do we lose? The answer is the extended list, and you see that here. If a graph is admissible, you will always get the right answer unless you use an extended list because you're checking from every node to the goal node. And you're sure that your estimates are right. But if estimates within nodes aren't correct, you might go through them out of order, and that violates your assumption that you made when you decided to use the extended list that you would always go through the sub-graphs in order. This graph is very expertly crafted to do that to you cause I might as well be sort of a bottleneck, goal node, and I has an inconsistency. There's a few other inconsistencies in this graph, but they don't do anything to you. In fact, even inadmissibilities sometimes don't mess you up. So you can't just say, blindly, it's inadmissible, so it will never work. It might work. It turns out the only inconsistency that matters here is not the inconsistency between, say, S and F. That doesn't turn out to matter. It's the inconsistency between some of these nodes, including I and H, specifically. So have a good weekend. Come in on Monday and Tuesday to tutorial. Ask about the queue method, or other methods. Ask about anything that's niggling in your brain. But hopefully, you guys have got this. You're going to do fine.
MIT_6034_Artificial_Intelligence_Fall_2010	21_Probabilistic_Inference_I.txt	PATRICK WINSTON: Here we are, down to the final sprint. Three to go. And we're going to take some of the last three, maybe two of the last three, to talk a little bit about stuff having to do with probabilistic approaches-- use of probability in artificial intelligence. Now, for many of you, this will be kind of a review, because I know many of you learned about probability over the [? sand ?] table and every year since then. But maybe we'll put another little twist into it, especially toward the end of the hour when we get into a discussion of that which has come to be called belief nets. But first, I was driving in this morning, and I was quite astonished to see, as I drove in, this thing here. And my first reaction was, oh my god, it's the world's greatest hack. And then I decided, well, maybe it's a piece of art. So I'd like to address the question of how I could come to grips with that issue. There's a distinct possibility that this thing is a consequence of a hat, possibly the result of some kind of art show. And in any event, some sort of statue appeared, and statues don't usually appear like that. So I got the possibility of thinking about how all these things might occur together or not occur together. So the natural thing is to build myself some sort of table to keep track of my observations. So I have three columns in my table. I've got the possibility of a statue appearing, a hack having occurred, and some sort of art show. And so I can make a table of all the combinations of those things that might appear. And I happen to have already guessed that there are going to be eight rows in my table. So it's going to look like this. And this is the set of combinations in this row where none of that occurs at all. And down here is the situation where all of those things occur. After all, it's possible that we can have an art show and have a hack be a legitimate participant in the art show. That's why we have that final row. So we have all manner of combinations in between. So those are those combinations. Then we have F, F, T, T, F, F, T, T, F, T, F, T, F, T, F, T. So it's plain that the number of rows in the table, or these binary possibilities, is 2 to the number of variables. And that could be a big number. In fact, I'd love to do a bigger example, but I don't have the patience to do it. But anyhow, what we might do is in order to figure out how likely any of these combinations are, is we might have observed the area outside the student center and rest of campus over a long period of time and keep track of what happens on 1,000 days. Or maybe 1,000 months or 1,000 years. I don't know. The trouble is, these events don't happen very often. So the period of time that I use for measurement needs to be fairly long. Probably a day is not short enough. But in any case, I can keep a tally of how often I see these various combinations. So this one might be, for example, 405, this one might be 45, this one might be 225, this one might be 40, and so on. And so having done all those measurements, kept track of all that data, then I could say, well, the probability that at any given time period one of these things occurs will just be the frequency-- the number of tallies divided by the total number of tallies. So that would be a number between 0 and 1. So that's the probability for each of these events. And it's readily calculated from my data. And once I do that, then I can say that I got myself a joint probability table, and I could perform all manner of miracles using that joint probability table. So let me perform a few of those miracles, while we're at it. There's the table. And now, what I want to do is I want to count up the probability in all the rows where the statue appears. So that's going to be the probability of the statue appearing. So I'll just check off those four boxes there. And it looks like the probability of the statue appearing is about 0.355 in my model. I don't think it's quite that frequent, but this is a classroom exercise, right? So I can make up whatever numbers I want. Now, I could say, well, what's the probability of a statue occurring given that there's an art show? Well, I can limit my tallies to those in which art show is true, like so. And in that case, the probability has just zoomed up. So if I know there's an art show, there's a much higher probability that a statue will appear. And if I know there's a hack as well as an art show going on, it goes up higher still to 0.9. We can also do other kinds of things. For example, we can go back to the original table. And instead of counting up the probability we've got a statue, as we just did, we're going to calculate the probability that there is an art show. I guess that would be that one and that one, not that one, but that one. So the probability there's an art show is one chance in 10. Or we can do the same thing with a hack. In that case, we get that one off, that one on, that one off, that one on, that one off, that one on, that one off. So the probability of a hack on any given time period is about 50-50. So I've cooked up this little demo so it does the "ands" of all these things. It could do "ors," too, with a little more work. But these are just the "ands" of these various combinations. Then you can ask more complicated questions, like for example, you could say, what is the probability of a hack given that there's a statue? And that would be limiting the calculations to those rows in which the statue thing is true. And then what I get is 0.781. Now, what would happen to the probability that it's a hack if I know that there's an art show? Will that number go up or down? Well, let's try it. Ah, it went down. So that's sort of because the existence of the art show sort of explains why the statue might be there. Now, just for fun, I'm going to switch to another situation, very similar. And the situation here is that a neighbor's dog often barks. It might be because of a burglar. It might be because of a raccoon. Sometimes, there's a burglar and a raccoon. Sometimes, the damn dog just barks. So let's do some calculations there and calculate the probability that a raccoon is true, similar to what we did last time. Looks like on any given night-- it's kind of a wooded are-- there's a high probability of a raccoon showing up. And then we can ask, well, what is the probability of the dog barking given that a raccoon shows up? Well, in that case, we want to just limit the number of rows to those where a raccoon-- or where the dog is barking. Looks like the probability of the dog barking, knowing nothing else, is about [? 3/7. ?] But now we want to know the probability of the raccoon-- that's these guys here need to get checked. These are off. So that's the probability of a raccoon. Did I get that right? Oh, that's probability of a burglar. Sorry, that was too hard. So let me go back and calculate-- I want to get the probability of a raccoon. That's true, false, true, false, true, false, true. So the probability of a raccoon, as I said before is 0.5. Now, what happens to that probability if I know the dog is barking? Well, all I need to do is limit my rows to those where the dog is barking, those bottom four. And I'll click that there, and you'll notice all these tallies up above the midpoint have gone to zero, because we're only considering those cases where the dog is barking. In that case, the probability that there's a raccoon-- just the number of tallies over the total number of tallies-- gee, I guess it's 225 plus 50 divided by 370. That turns out to be 0.743. So about 75% of the time, the dog barking is accounted for-- well, the probability of a raccoon under those conditions is pretty high. And now, once again, I'm going to ask, well, what is the probability of a raccoon, given that the dog is barking and there's a burglar? Any guess what will happen there? We did this once before with the statue. Probability first went up when we saw the statue and then went down when we saw another explanation. Here's this one here. Wow, look at that. It went back to its original condition, its a priori probability. So somehow, the existence of the burglar and the dog barking means that the probability of a raccoon is just what it was before we started this game. So those are kind of interesting questions, and there's a lot we can do when we have this table by way of those kinds of calculations. And in fact, the whole miracle of probabilistic inference is right in front of us. It's the table. So why don't we go home? Well, because there's a little problem with this table-- with these two tables that I've shown you by way of illustration. And the problem is that there are a lot of rows. And I had a hard time making up those numbers. I didn't have the patience to wait and make observations. That would take too long. So I had to kind of make some guesses. And I could kind of manage it with eight rows-- those up there. I could put in some tallies. It wasn't that big of a deal. So I got myself all those eight numbers up there like that. And similarly, for the art show calculations, produced eight numbers. But what if I added something else to the mix? What if I added the day of the week or what I had for breakfast? Each of those things would double the number of rows of their binary variables. So if I have to consider 10 influences all working together, then I'd have 2 to the 10th. I'd have 1,000 numbers to deal with. And that would be hard. But if I had a joint probability table, then I can do these kinds of miracles. But Dave, if I could have this little projector now, please. I just want to emphasize that although we're talking about probabilistic inference, and it's a very powerful tool, it's not the only tool we need in our bag. Trouble with most ideas in artificial intelligence is that their hardcore proponents think that they're the only thing to do. And probabilistic inference has a role to play in developing a theory of human intelligence. And it certainly has a practical value, but it's not the only thing. And to illustrate that point, I'd like to imagine for a few moments that MIT were founded in 1861 BC instead of 1861 AD. And if that were so, then it might be the case that there would be a research program on what floats. And this, of course, would be a problem in experimental physics, and we could imagine that those people back there in that early MIT would, being experimentally minded, try some things. Oh, I didn't know that's what happened. It looks like chalk floats. Here's a rock. No, it didn't float. Here's some money. Doesn't float. Here's a pencil. No, it doesn't float. Here's a pen. Here's a piece of tin foil I got from Kendra. That floats. That's a metal. The other stuff's metal, too. Now I'm really getting confused. Here's a little wad of paper. Here's a cell ph-- no, actually, I've tried that before. They don't float. And they also don't work afterward, either. I don't need to do any of that in the MIT of 1861 AD and beyond, because I know that Archimedes worked this all out. And all I have to do is measure the volume of the stuff, divide that by the weight, and if that ratio is big enough, then the thing will float. But back in the old days, I would have to try a lot of stuff and make a big table, taking into account such factors as how hard it is, how big it is, how heavy it is, whether it's alive or not. Most things that are alive float. Some don't. Fish don't, for instance. So it would be foolhardy to do that. That's sort of a probabilistic inference. On the other hand, there are lots of things where I don't know all the stuff I need to know in order to make the calculation. I know all the stuff I need to know in order to decide if something floats, but not all the stuff I need to know in order, for example, to decide if the child of a Republican is likely to be a Republican. There are a lot of subtle influences there, and it is the case that the children of Republicans and the children of Democrats are more likely to share the political party of their parents. But I don't have any direct way of calculating whether that will be true or not. All I can do in that case is what I've done over here, is do some measurements, get some frequencies, take some snapshots of the way the world is and incorporate that into a set of probabilities that can help me determine if any given parent is a Republican, given that I've observed the voting behavior their children. So probability has a place, but it's not the only tool we need. And that is an important preamble to all the stuff we're going to do today. Now, we're really through, because this joint probability table is all that there is to it, except for the fact we can't either record all those numbers, and it becomes quickly a pain to guess at them. There are two ways to think about all this. We can think about these probabilities as probabilities that come out of looking at some data. That's a frequentist view of the probabilities. Or we could say, well, we can't do those measurements. So I can just make them up. That's sort of the subjective view of where these probabilities come from. And in some cases, some people like to talk about natural propensities, like in quantum mechanics. But for our purposes, we either make them up, or we do some tallying. Trouble is, we can't deal with this kind of table. So as a consequence of not being able to deal with this kind of table, a gigantic industry has emerged for dealing with probabilities without the need to work up this full table. And that's where we're going to go for the rest of the hour. And here's the path we're going to take. We're going to talk about some basic overview of basic probability. Then we're going to move ourselves step by step toward the so-called belief networks, which make it possible to make this a practical tool. So let us begin. The first thing is basic probability. Let us say basic. And basic probability-- all probability flows from a small number of axioms. We have the probability of some event a has got to be greater than 0 and less than 1. That's axiom number one. In a binary world, things have a probability of being true. Some have a probability of being false. But the true event doesn't have any possibility of being anything other than true, so the probability of true is equal to 1, and the probability of false-- the false event, the false condition-- has no possibility of being true, so that's 0. Then the third of the axioms of probability is that the probability of a plus the probability of b minus the probability of a and b is equal to the probability of a or b. Yeah, that makes sense, right? I guess it would make more sense if I didn't switch my notation in midstream-- a and b. So those are the axioms that mathematicians love to start up that way, and they can derive everything there is to derive from that. But I never can deal with stuff that way. I have to draw a picture and think of this stuff in a more intuitionist type of way. So that's the formal approach to dealing with probability, and it's mirrored by intuitions that have to do with discussions of spaces, like so, in which we have circles, or areas, representing a and b. And to keep my notation consistent, I'll make those lowercase. So you can think of those as spaces of all possible worlds in which these things might occur. Or you can think of them as sample spaces. But in any event, you associate with the probability of a the size of this area here relative to the total area in the rectangle-- the universe. So the probability of a is the size of this circle divided by the size of this rectangle in this picture. So now all these axioms make sense. The probability that a is certain is just when that fills up the whole thing, and there's no other place for a sample to be, that means it has to be a. So that probability goes all the way up to 1. On the other hand, if the size of a is just an infinitesimal dot, then the chances of landing in that world is 0. That's the bound on the other end. So this-- axiom number one-- makes sense in terms of that picture over there. Likewise, axiom number two. What about axiom number three? Does that make sense in terms of all this stuff? And the answer is, sure, because we can just look at those areas with a little bit of colored chalk. And so the probability of a is just this area here. The probability of b is this area here. And if we want to know the probability that we're in either a or b, then we just have to add up those areas. But when we add up those areas, this intersection part is added in twice. So we've got to subtract that off in order to make this thing make a rational equation, so that makes sense. And axiom three makes sense, just as axioms one and two did. So that's all there is to basic probability. And now you could do all sorts of algebra on that, and it's elegant, because it's like circuit theory or electromagnetism, because from a very small number of axioms-- in this case three-- you can build an elegant mathematical system. And that's what probability subjects do. But we're not going to go there, because we're sort of focused on getting down to a point where we can deal with that joint probability table that we currently can't deal with. So we're not going to go into a whole lot of algebra with these things. Just what we need in order to go through that network. So the next thing we need to deal with is conditional probability. And whereas those are axioms, this is a definition. We say that the probability of a given b is equal to, by definition, the probability of a and b. I'm using that common notation to mean [INAUDIBLE] as is conventional in the field. And then we're going to divide that by the probability of B. You can take that as a definition, and then it's just a little bit of mysterious algebra. Or you could do like we did up there and take an intuitionist approach and ask what that stuff means in terms of a circle diagram and some sort of space. And let's see, what does that mean? It means that we're trying to restrict the probability of a to those circumstances where b is known to be so. And we're going to say that-- we've got this part here, and then we've got the intersection of a with b. And so it does make sense as a definition, because it says that if you've got b, then the probability that you're going to get a is the size of that intersection-- the pink and orange stuff-- divided by the whole of b. So it's as if we restricted the universe of consideration to just that part of the original universe as covered by b. So that makes sense as a definition. And we can rewrite that, of course, as P of a and b is equal to the probability of a given b times the probability of b. That's all basic stuff. Now, we do want to do a little bit of algebra here, because I want to consider not just two cases, but what if we divide this space up into three parts? Then we'll say that the probability of a, b, and c is equal to what? Well, there are lots of ways to think about that. But one way to think about it is that we are restricting the universe to that part of the world where b and c are both true. So let's say that y is equal to b and c-- the intersection of b and c, where a and b are both true. Then we can use this formula over here to say that probability of a, b, and c is equal to the probability of a and y, which is equal to the probability of a given y times the probability of y. And then we can expand that back out and say that P of a given b and c is equal to the probability-- sorry, times the probability of y, but y is equal to the probability of b and c, like so. Ah, but wait-- we can run this idea over that one, too, and we can say that this whole works is equal to the probability of a given b and c times the probability of b given c times the probability of c. And now, when we stand back and let that sing to us, we can see that some magic is beginning to happen here, because we've taken this probability of all things being so, and we've broken up into a product of three probabilities. The first two are conditional probabilities, so they're really all conditional probabilities. The last one's conditional on nothing. But look what happens as we go from left to right. a is dependent on two things. b is only dependent on one thing and nothing to the left. c is dependent on nothing and nothing to the left. So you can sense a generalization coming. So let's write it down. So let's go from here over to here and say that the probability of a whole bunch of things-- x1 through x10-- is equal to some product of probabilities. We'll let the index i run from n to 1. Probability of x to the last one in the series, conditioned on all the other ones-- sorry, that's probability of i, i minus 1 down to x1 like so. And for the first one in this product, i will be equal to n. For the second one, i will be equal to n minus 1. But you'll notice that as I go from n toward 1, these conditionals get smaller-- the number of things on condition get smaller, and none of these things are on the left. They're only stuff that I have on the right. So what I mean to say is all of these things have an index that's smaller than this index. None of the ones that have a higher index are appearing in that conditional. So it's a way of taking a probability of the end of a whole bunch of things and writing it as a product of conditional probabilities. So we're making good progress. We've done one. We've done two. And now we've done three, because this is the chain rule. And we're about halfway through our diagram, halfway to the point where we can do something fun. But we still have a couple more concepts to deal with, and the next concept is the concept of conditional probability. So that's all this stuff up here-- oops. All this stuff here is the definition of conditional probability. And now I want to go to the definition of independence. So that's another definitional deal. But it's another definitional deal that makes some sense with a diagram as well. So the definition goes like this. We say that P of a given b is equal to P of a if a independent of b. So that says that the probability of a doesn't depend on what's going on with b. It's the same either way. So it's independent. b doesn't matter. So what does that look like if we try to do an intuitionist diagram? Well, let's see. Here's a. Here's b. Now, the probability of a given b-- well, let's see. That must be this part here divided by this part here. So the ratio of those areas is the probability of a given b. So that's the probability of this way divided by the probability of both ways. So what's the probability of a in terms of these areas? Well, probability of a in terms of these areas is the probability-- let's see, have I got this right? I've got this upside down. The probability of a given b is the probability of the stuff in the intersection-- so that's both ways-- divided by the probability of the stuff in b, which is going this way. And let's see, the probability of a not conditioned on anything except being in this universe is all these hash marks, like so, divided by the universe. So when we say that something's independent, it means that those two ratios are the same. That's all it means in the intuitionist's point of view. So it says that this little area here divided by this whole area is the same as this whole area for a divided by the size of the universe. So that's what independence means. Now, that's quite a lot of work. But we're not done with independence, because we've got conditional independence to deal with. And that, too, can be viewed as a definition. And what we're going to say is that the probability of a given b and z is equal to the probability of a given z. What's that mean? That means that if you know that we're dealing with z, then the probability of a doesn't depend on b. b doesn't matter anymore once you're restricted to being in z. So you can look at that this way. Here's a, and here's b, and here is z. So what we're saying is that we're restricting the world to being in this part of the universe where z is. So the probability of a given b and z is this piece in here. a given b and z is that part there. And the probability of a given z is this part here divided by all of z. So we're saying that the ratio of this little piece here to this part, which I'll mark that way, ratio of this to this is the same as the ratio of that to that. So that's conditional independence. So you can infer from these things, with a little bit of algebra, that P of a and b given z is equal to P of a given z times P of b in z. Boy, that's been quite a journey, but we got all the way through one, two, three, four, and five. And now the next thing is belief nets, and I'm going to ask you to forget everything I've said for a minute or two. And we'll come back to it. I want to talk about the dog and the burglar and the raccoon again. And now, forgetting about probability, I can say, look, the dog barks if a raccoon shows up. The dog barks if a burglar shows up. A burglar doesn't show up because the dog is barking. A raccoon doesn't show up because the dog is barking. So the causality flows from the burglar and the raccoon to the barking. So we can make a diagram of that. And our diagram will look like this. Here is the burglar, and here is the raccoon. And these have causal relations to the dog barking. So that's an interesting idea, because now I can say that-- well, I can't say anything yet, because I want to add a little more complexity to it. I'm going to add two more variables. You might call the police, depending on how vigorous the dog is barking, I guess. And the raccoon has a propensity to knocking over the trash can. So now, I've got five variables. How big a joint probability table am I going to need to keep my tallies straight? Well, it'll be 2 to the 5th. That's 32. But what I'm going to say is that this diagram is a statement, that every node in it depends on its parents and nothing else that's not a descendant. Now, I need to say that about 50 times, because you've got to say it right. Every node there is independent of every non-descendant other then its parents. No, that's not quite right. Given its parents, every node is independent of all other non-descendants. Well, what does that mean? Here's the deal with calling the police. Here's its one and only parent. So given this parent, the probability that they were going to call the police doesn't depend on anything like B, R, or T. It's because all of the causality is flowing through this dog barking. I'm not going to call the police in a way that's dependent on anything else other than whether the dog is barking or not. Because this guy has this as a parent, and these are not descendants of calling the police, so this is independent of B, R, and T. So let's go walk through the others. Here's the dog. The dog's parents are burger appearing and raccoon appearing. So the probability that the dog appears is independent of that trash can over there, because that's not a descendant. It is dependent on these parents. How about the trash can? It depends only on the raccoon. It doesn't depend on any other non-descendant, so therefore, it doesn't depend on D, B, or P. How about B? It has no parents. So it depends on nothing else, because everything else is either a non-descendant, because B does not dependent on R and T, because they're not descendants. It's interesting that B might depend on D and P, because those are descendants. So it's important to understand that there's this business of independence given the parents of all other non-descendants. And you'll see why that funny, strange language is important in a minute. But now, let's see-- I want to make a model of what's going to happen here. So let me see what kind of probabilities I'm going to have to figure out. This guy doesn't depend on anything upstream. So we could just say that all we need there is the probability that a burglar is going to appear. Let's say it's a fairly high-crime neighborhood-- 1 chance in 10-- 1 day in 10, a burglar appears. The raccoon doesn't depend on anything other than its own propensity, so its probability, we'll say, is 0.5. Raccoons love the place, so it shows up about 1 day in 2. So what about the dog barking? That depends on whether there's a burglar, and the other parent is whether there's a raccoon. So we need to keep track of the probability that the dog will bark for all four combinations. So this will be the burglar, and this will be the raccoon. This will be false, false, true, true-- oops-- false, false, true, false, false, true, true, true. So let's say it's a wonderful dog, and it always barks if there's a burglar. So that would say that the probability here is 1.0, and the probability here is 1.0. And if there's neither a burglar nor a raccoon, the dog still likes to bark just for fun. So we'll say that's a chance of 1 in 10. And then in case there's a burglar, let's say this. There's no burglar, but there is a raccoon-- he's tired of the raccoons, so he only barks half the time. Do these numbers, by the way, have to add up to 1? They clearly don't. These numbers don't add up to one. What adds up to 1 is this is the probability that the dog barks. And then the other phantom probability is out here. And these have to add up to 1. So that would be 0.9, that would be 0.0, that would be 0.5, and this would be 0.0. So because those are just 1 minus the numbers in these columns, I don't bother to write them down. Well, we still have a couple more things to do. The probability that we'll call the police depends only on the dog. So we'll have a column for the dog, and then we'll have a probability of calling the police. There's a probability for that being false and a probability for that being true. So if the dog doesn't bark, there's really hardly any chance we'll call the police. So make that 0, 0, 1. If the dog is barking, if he barks vigorously enough, maybe 1 chance in 10. Here, we have the trash can-- the final thing we have to think about. There's the trash can; rather, the raccoon. And here's the trash can probability. Depends on the raccoon being either present or not present. If the raccoon is not present, the probability the trash can is knocked over by, say, the wind is 1 in 1,000. If the raccoon is there, oh man, that guy always likes to go in there, so that's 0.8. So now I'm done specifying this model. And the question is, how many numbers did I have to specify? Well, let's see. I have to specify that one, that one, that one, that one, that one, that one-- that's 6, 7, 8, 9, 10. So I had to specify 10 numbers. If I just try to build myself a joint probability table straightaway, how many numbers would I have to supply? Well, it's 2 to the n. So it's 2 to the 5th, that's 32. Considerable saving. By the way, how do you suppose I made that table? Not by doing all those numbers. By making this belief network and then using the belief network to calculate those numbers. And that's why this is a miracle, because with these numbers, I can calculate those numbers instead of making them up or making a whole lot of tally-type measurements. So I'd like to make sure that that's true. And I can use this stuff here to calculate the full joint probability table. So here's how this works. I have the probability of some combination-- let's say the police, the dog, the burglar, the trash can, and the raccoon. All the combinations that are possible there will give me an entry in the table-- one row. But let's see-- there's some miracle here. Oh, this chain rule. Let's use the chain rule. We can write that as a probability that we call the police given d, b, t, and r. And then the next one in my chain is probability of d given b, t, and r. Then the next one in the chain is the probability of b given t and r. And the next one in my chain is P of t given r. And the final one in my chain is p of r. Now, we have some conditional independence knowledge, too, don't we? We know that this probability here depends only on d because there are no descendants. So therefore, we don't have to think about that, and all the numbers we need here are produced by this table. How about this one here? Probability that the dog barks depends only on its parents, b and r, so it doesn't depend on t. So b, in turn, depends on-- what does it depend on? It doesn't depend on anything. So we can scratch those. Probability of t given r, yeah, there's a probability there, but we can get that from the table. And finally, P or r. So that's why I went through all that probability junk, because if we arrange things in the expansion of this, from bottom to top, then we arrange things so that none of these guys depends on a descendant in this formula. And we have a limited number of things that it depends on above it. So that's the way we can calculate back the full joint probability table. And that brings us to the end of the discussion today. But the thing we're going to think about is, how much saving do we really get out of this? In this particular case, we only had to devise 10 numbers out of 32. What if we had 10 properties or 100 properties? How much saving would we get then? That's what we'll take up next time, after the quiz on Wednesday.

Subsets and Splits

No community queries yet

The top public SQL queries from the community will appear here once available.